A method for detecting anomalies in bridge deflection monitoring response based on a deep convolutional autoencoder

Abstract

This invention discloses a method for detecting anomalies in bridge deflection monitoring response based on a deep convolutional autoencoder, comprising: collecting bridge deflection monitoring responses over a time period t. D ( t An ideal filter separates components based on the lower and upper frequency cutoff limits. D 1( t ), D 2( t )and D 3( t Using the moving average method to select the window D 2( t Separate into D 21 ( t )and D 22 ( t To construct a sample dataset of bridge deflection monitoring response under normal conditions, the aforementioned deflection response components under normal conditions were obtained using the method described above. A deep convolutional network (DCNN) was designed, and the time-history response images of the deflection response components were used as input to train four information feature extraction models. A classification model was constructed by reconstructing the error, and the measured deflection response DCNN was then used to classify the data. m (t) The measured deflection response components are obtained according to the above method and input into the information feature extraction model respectively. The reconstruction error is used to classify them into normal state and abnormal state respectively.

Description

Technical Field

[0001] This invention relates to the field of bridge deformation measurement technology, and more specifically, to a method for detecting abnormal bridge deflection monitoring response based on a deep convolutional autoencoder. Background Technology

[0002] As of now, China has over 1 million highway bridges, and the number of bridges built continues to grow rapidly each year. The number of freight vehicles has reached 11.6666 million, with a total tonnage of 169.6733 million tons and an average freight vehicle load of 14.5435 tons. This demonstrates that highway transportation infrastructure continues to bear the heavy burden of heavy-duty transportation, which exposes many bridges that have been in service for many years to operational safety risks. With increasing traffic volume and the aging of bridges, bridge structures may suffer damage and degradation due to environmental erosion, traffic load, and material aging. If this damage is not detected and repaired in a timely manner, it could affect driving safety, shorten the bridge's lifespan, or even lead to sudden damage and collapse.

[0003] Currently, the most common method for bridge inspection is radar interferometry, which achieves millimeter-level accuracy. This technique allows for high-precision measurement of bridge deflection, reflecting the bridge's deformation. However, when vehicles pass over the bridge or factors such as wind influence it, the bridge deflection changes dynamically. This dynamic change affects radar interferometry, leading to inaccurate measurements of bridge deformation based on the measured deflection.

[0004] Therefore, there is an urgent need for a bridge health monitoring system to inspect the condition of bridges at different times and under various conditions, improve inspection efficiency and accuracy, provide real-time data on the health status of bridge structures, and help engineers make better maintenance decisions. Monitoring systems typically include various sensors, such as displacement sensors, strain sensors, accelerometers, and temperature sensors, to monitor key bridge parameters, such as vertical acceleration and deflection. Summary of the Invention

[0005] The present invention aims to overcome at least one defect in the prior art and provide a bridge deflection monitoring response anomaly detection method based on a deep convolutional autoencoder, which is used to solve the problem of inaccurate bridge deflection measurement when affected by factors such as vehicles or wind.

[0006] The technical method adopted in this invention is a bridge deflection monitoring response anomaly detection method based on a deep convolutional autoencoder, comprising the following steps:

[0007] Step 100: Collect bridge deflection monitoring response over a time period t. D ( t );

[0008] Step 200: Design a series of ideal filters and set the lower frequency cutoff limit according to the characteristics of the bridge. w l and frequency cutoff limit w u Ideal filters are based on w l and w u Components separated D 1( t ), D 2( t )and D 3( t );

[0009] Step 300: Select the window using the moving average method. D 2( t Separate into D 21 ( t )and D 22 ( t );

[0010] Step 400: Constructing the bridge deflection monitoring response under normal conditions D ( t Using the sample dataset, the deflection response components under normal conditions were obtained according to the method described above. D 1( t ), D 21 ( t ), D 22 ( t )and D 3( t );

[0011] Step 500: Design a deep convolutional network (DCNN) to... D 1( t ), D 21 ( t ), D 22 ( t )and D 3( t Using the time-series response images as input, the information feature extraction models DCNN1 and DCNN are trained. 21 DCNN 22 And DCNN3, which constructs a classification model by reconstructing errors;

[0012] Step 600: Calculate the measured deflection response D m(t) The four components obtained by the separation method described above D m 1( t ), D m 21 ( t ), D m 22 ( t )and D m 3( t Input the trained DCNN1 and DCNN respectively. 21 DCNN 22 In DCNN3, the reconstruction error is used to classify the state into normal and abnormal states.

[0013] Bridge deflection monitoring response data over a period of time (t) are collected using high-precision sensors. D ( t This ensures the accuracy and completeness of the data; an ideal filter is designed based on the structural characteristics and dynamic behavior of the bridge, and a frequency cutoff upper limit is set. w u and frequency cutoff lower limit w l The low-frequency response of the bridge deflection response is separated by a filter. D 1( t ), intermediate frequency response D 2( t and high frequency response D 3( t This enables refined signal processing.

[0014] To gain a deeper understanding of the frequency characteristics of the bridge deflection response, step 100 specifically involves collecting the bridge deflection monitoring response over a time period t. D ( t ), obtain spectral information through Fast Fourier Transform. F ( w The Fast Fourier Transform (FFT) technique can decompose complex time-domain signals into a series of simple sine or cosine waves, each corresponding to a specific frequency component. By converting the response into spectral information, the distribution of the bridge deflection response at different frequencies can be reflected. By observing and analyzing the spectral information F(w), the dominant frequencies, secondary frequencies, and potential anomalous frequency components in the bridge deflection response can be identified, providing strong support for subsequent signal separation, feature extraction, and anomaly detection.

[0015] To separate signal components, the ideal filter in step 200 is based on... w l andw u Components separated D 1( t ), D 2( t )and D 3( t Specifically, the frequency w < w l The components are separated into [a specific composition] using an ideal filter. F 1( w ), frequency w l ≤ w ≤ w u The components are separated into [a specific composition] using an ideal filter. F 2( w ), frequency w > w u The components are separated into [a specific composition] using an ideal filter. F 3( w ), by inverse Fourier transform, the components F 1( w ), F 2( w )and F 3( w They were converted into time domain signals respectively, and the results were obtained respectively. D 1( t ), D 2( t )and D 3( t ). Regarding frequency w < w l The components are analyzed, and a low-pass ideal filter is designed to separate the low-frequency components from the original spectral information, forming new spectral components. F 1( w For frequency w l ≤ w ≤ w u The components are determined by designing an ideal bandpass filter to extract the signal within this frequency range, thus forming the spectral components. F 2( w ); Regarding frequency w > w u The components are analyzed, and a high-pass ideal filter is designed to separate the high-frequency noise components from the original spectral information, forming the spectral components. F 3( w After completing the fine separation of the spectral components, the spectral components are then analyzed using inverse Fourier transform.F 1( w ), F 2( w )and F 3( w The signals are converted back to the time domain, preserving the phase information of the original signals while also ensuring the stability of the time domain signals. D 1( t ), D 2( t )and D 3( t The accuracy and completeness of ( ).

[0016] The D 1( t )+ D 21 ( t The ) represents the low-frequency pseudo-static components caused by long-term downward deflection, annual temperature difference, monthly temperature difference, and daily temperature difference, while the ) represents the mid-frequency pseudo-static components caused by average wind field and traffic congestion. D 22 ( t ) represents the high-frequency pseudo-static component caused by sparse traffic, the D 3( t This represents the high-frequency dynamic components caused by traffic flow dynamics, environmental vibrations, and noise. The high-frequency quasi-static component is identified because the irregularity and instantaneity of traffic flow result in more frequent and rapid fluctuations, thus exhibiting high-frequency quasi-static characteristics in dynamic analysis. High-frequency dynamic components are not only high in frequency and rapid in change, but are also often accompanied by significant energy transfer and dynamic response, which have a significant impact on the overall stability and safety of the structure.

[0017] To efficiently process complex data, the Deep Convolutional Network (DCNN) in step 500 includes an encoder module, a spatial-channel self-attention module, and a decoding module. The encoder module extracts low-dimensional feature representations of the input data through multiple convolutional layers. The spatial-channel self-attention module utilizes an attention mechanism to dynamically adjust feature weights in both spatial and channel dimensions, thereby capturing and enhancing key information and improving the model's ability to recognize complex feature patterns. The decoding module maps the features processed by the encoding and self-attention mechanisms back to the original data space, and through upsampling and deconvolution operations, gradually restores the spatial resolution and detailed information of the data, ultimately outputting high-quality prediction or classification results.

[0018] To achieve feature extraction and dimensionality reduction, the encoder module includes four convolutional sub-modules, each comprising a convolutional layer, a downsampling layer, and an activation function. The convolutional layers, through multi-kernel convolution operations, delve into the local adjustments of the input data to generate feature maps, providing crucial information for subsequent processing. The downsampling layers, through pooling operations (such as max pooling or average pooling), reduce the size of the feature maps, lowering the data dimensionality while preserving important feature information, thus enhancing the model's robustness and generalization ability. The activation function introduces a non-linear activation function, introducing non-linear characteristics to the model, enabling the network to learn and represent complex non-linear relationships, thereby improving the model's expressive power.

[0019] To progressively recover and refine feature map information, the decoding module includes three deconvolution sub-modules, each comprising a deconvolution layer, an upsampling layer, and an activation function. The deconvolution layer uses deconvolution operations (transposed convolution) to gradually map the feature map from low resolution back to high resolution, simultaneously extracting and enhancing feature information to provide data for subsequent image reconstruction or feature recovery. The upsampling layer employs upsampling algorithms (such as bilinear interpolation, nearest neighbor interpolation, or upsampling networks) to increase the size of the feature map, recovering image details while maintaining feature consistency and coherence. The activation function is a non-linear activation function (such as ReLU, Leaky ReLU, PReLU, or ELU) to introduce non-linear characteristics into the decoding process, enhancing the model's expressive power and making the decoded image or features more realistic and natural.

[0020] The condition for determining the normal state in step 600 is that when all four classification modules produce normal results, the deflection response is determined to be in a normal state.

[0021] To distinguish between different types of anomalies, the criteria for determining the anomaly state in step 600 are as follows:

[0022] (1) When DCNN1, DCNN 21 DCNN 22 The existence of anomalies in DCNN1 and DCNN2, respectively, is considered as anomaly states 1-4. That is, anomaly state 1 is an anomaly in DCNN1, anomaly state 2 is an anomaly in DCNN2, and anomaly state 3 is an anomaly in DCNN2. 21 An anomaly has occurred; anomaly state 3 is DCNN. 22 An anomaly has occurred; anomaly status 4 indicates that DCNN3 has encountered an anomaly.

[0023] (2) When DCNN1, DCNN 21 DCNN 22 The combination of DCNN1 and DCNN3 exhibits anomalies, which are considered anomalous states 5-10. Anomalous state 5 refers to the combination of DCNN1 and DCNN3. 21 All showed abnormalities, with abnormal state 6 being DCNN1 and DCNN. 22All showed anomalies; anomaly state 7 indicated that both DCNN1 and DCNN3 showed anomalies; anomaly state 8 indicated that DCNN... 21 and DCNN 22 All showed abnormalities, with abnormal state 9 being DCNN. 21 Both DCNN and DCNN3 exhibited anomalies, with anomaly state 10 being DCNN. 22 Both DCNN and DCNN3 showed anomalies;

[0024] (3) When DCNN1, DCNN 21 DCNN 22 The combination of DCNN1, DCNN2, and DCNN3 exhibits anomalies, which are considered anomalous states 11-14. Specifically, anomalous state 11 refers to DCNN1 and DCNN3. 21 and DCNN 22 All showed abnormalities, with abnormal state 12 being DCNN1 and DCNN. 21 Both DCNN1 and DCNN3 exhibited anomalies, while anomaly state 13 was observed in DCNN1 and DCNN3. 22 Both DCNN and DCNN3 exhibited anomalies, with anomaly state 14 being DCNN. 21 DCNN 22 Both DCNN and DCNN3 showed anomalies;

[0025] (4) When DCNN1, DCNN 21 DCNN 22 Anomaly 15 occurs when all four modules (DCNN, DCNN3, and DCNN3) are abnormal. When anomaly 15 occurs, it means that all modules are abnormal and emergency intervention is required.

[0026] The 15 abnormal states correspond to the following abnormal types: Abnormal state 1 corresponds to abnormal temperature difference; Abnormal state 2 corresponds to abnormal interference and abnormal noise; Abnormal state 3 corresponds to data loss and low resolution; Abnormal state 4 corresponds to abnormal overloading; Abnormal state 5 corresponds to abnormal temperature difference, abnormal interference, and abnormal noise; Abnormal state 6 corresponds to abnormal temperature difference, data loss, and low resolution; Abnormal state 7 corresponds to abnormal temperature difference and abnormal overloading; Abnormal state 8 corresponds to abnormal interference, abnormal noise, data loss, and low resolution; Abnormal state 9 corresponds to abnormal interference, abnormal noise, and abnormal overloading; Abnormal state 10 corresponds to data loss, low resolution, and abnormal overloading; Abnormal state 11 corresponds to abnormal temperature difference, abnormal noise, data loss, and low resolution; Abnormal state 12 corresponds to abnormal temperature difference, abnormal interference, abnormal noise, and abnormal overloading; Abnormal state 13 corresponds to abnormal temperature difference, data loss, low resolution, and abnormal overloading; Abnormal state 14 corresponds to abnormal interference, abnormal noise, data loss, low resolution, and abnormal overloading; and Abnormal state 15 corresponds to abnormal temperature difference, abnormal interference, abnormal noise, data loss, abnormal overloading, and sensor failure.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] 1. Real-time monitoring to determine anomaly type

[0029] The above methods enable real-time monitoring of abnormal bridge deflection data. By identifying anomalies in different frequency components of the data, it is possible to determine actual situations such as data loss, abnormal interference, abnormal heavy vehicle loads, abnormal temperature differences, abnormal noise, sensor malfunctions, and low resolution. This provides key methodological decisions for data analysis and abnormal state assessment of bridge deflection response health monitoring, which is of great significance for ensuring the safe operation of bridges. Attached Figure Description

[0030] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. To more clearly illustrate the technical methods of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0031] Figure 1 This is a flowchart of a bridge deflection monitoring response anomaly detection method based on a deep convolutional autoencoder according to the present invention.

[0032] Figure 2 This is a technical roadmap for a bridge deflection monitoring response anomaly detection method based on a deep convolutional autoencoder, according to the present invention.

[0033] Figure 3 This is an architectural diagram of a deep convolutional autoencoder according to the present invention.

[0034] Figure 4 The diagram shows the deflection response D(t) and spectrum of the bridge deflection monitoring response under normal conditions according to the present invention.

[0035] Figure 5 The diagram shows the deflection diagram and spectrum of the bridge deflection monitoring response Dm(t) under abnormal conditions according to the present invention. Detailed Implementation

[0036] The technical methods of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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.

[0037] Example 1

[0038] like Figure 1 and Figure 2As shown in the figure, this embodiment provides a method for detecting anomalies in bridge deflection monitoring response based on a deep convolutional autoencoder, including:

[0039] Step 100: Collect bridge deflection monitoring response over a time period t. D ( t );

[0040] Step 200: Design a series of ideal filters and set the lower frequency cutoff limit according to the characteristics of the bridge. w l and frequency cutoff limit w u , frequency w < w l The components are separated into [a specific composition] using an ideal filter. D 1( t ), frequency w l ≤ w ≤ w u The components are separated into [a specific composition] using an ideal filter. D 2( t ), frequency w > w u The components are separated into components using an ideal filter. D 3( t );

[0041] Step 300: Further employ the moving average method and select an appropriate window size. D 2( t Separate into D 21 ( t )and D 22 ( t ), respectively representing D 2( t Quasi-static and dynamic components;

[0042] Step 400: Constructing the bridge deflection monitoring response under normal conditions D ( t The sample dataset was used to obtain the deflection response components under normal conditions using the methods described above. D 1( t ), D 21 ( t ), D 22 ( t )and D 3( t );

[0043] Step 500: Design a deep convolutional autoencoder architecture, including an encoder module, a spatial-channel self-attention module, and a decoder module. D 1( t ), D 21 ( t ), D 22 ( t )or D 3( t Using time-response images as input, models DCNN1 and DCNN2 are trained to extract information features from the time-response images. 21 DCNN 22 And DCNN3, which constructs a classification model by reconstructing errors;

[0044] Step 600: Measure the deflection response D m ( t The four components obtained by the above separation method D m 1( t ), D m 21 ( t ), D m 22 ( t )and D m 3( t Input the trained DCNN1 and DCNN respectively. 21 DCNN 22 In DCNN3, the reconstruction error is used to classify the data into two states: normal and abnormal.

[0045] In this embodiment, step 100 specifically involves collecting the bridge deflection monitoring response over a time period t. D ( t ), obtain spectral information through Fast Fourier Transform. F ( w ).

[0046] Step 200 includes step 210 separating the spectral components and step 220 time-domain signal conversion.

[0047] Step 210: Separate the spectral components, specifically, various filters are applied to the components obtained in step 100. F ( w ) to perform filtering, the low-pass filter will filter the frequency w < w lThe components were separated into spectral components. F 1( w The bandpass filter will reduce the frequency. w l ≤ w ≤ w u The components were separated into spectral components. F 2( w The high-pass filter will reduce the frequency. w > w u The components were separated into spectral components. F 3( w ).

[0048] Step 220: Time-domain signal conversion, specifically, using inverse Fourier transform to convert the three components... F 1( w ), F 2( w )and F 3( w They were converted into time domain signals respectively, and the results were obtained respectively. D 1( t ), D 2( t )and D 3( t ).

[0049] Step 400: Construct the bridge deflection monitoring response under normal conditions. D ( t The sample dataset was used to obtain the deflection response components under normal conditions using the methods described above. D 1( t ), D 21 ( t ), D 22 ( t )and D 3( t Specifically, D 1( t )+ D 21 ( t The first part represents the low-frequency pseudo-static components caused by long-term downward deflection, annual temperature range, monthly temperature range, and daily temperature range, while the second part represents the mid-frequency pseudo-static components caused by mean wind field and traffic congestion. D 22 ( t ) represents the high-frequency pseudo-static component caused by sparse traffic. D 3( t This represents the high-frequency dynamic components caused by traffic flow dynamics, environmental vibrations, and noise.

[0050] In this embodiment, as Figure 3 As shown, step 500 involves designing a deep convolutional autoencoder architecture, comprising an encoder module, a spatial-channel self-attention module, and a decoder module. The encoder module includes four convolutional sub-modules, each containing a convolutional layer, a downsampling layer, and an activation function. These sub-modules extract image features from the image. The convolutional layers, through multi-kernel convolution operations, delve into the local adjustments of the input data to generate feature maps, providing crucial information for subsequent processing. The downsampling layers, through pooling operations (such as max pooling or average pooling), reduce the size of the feature maps, lowering the data dimensionality while retaining important feature information, thus enhancing the model's robustness and generalization ability. The activation function introduces a non-linear activation function, introducing non-linear characteristics to the model, enabling the network to learn and represent complex non-linear relationships, thereby improving the model's expressive power. The spatial-channel self-attention module utilizes an attention mechanism to dynamically adjust feature weights in both spatial and channel dimensions, thereby capturing and strengthening key information. This allows the network to focus more on images and image regions in a specific channel, enhancing its ability to identify image mutations and improving the model's ability to recognize complex feature patterns. The decoder module includes three deconvolution modules, each containing a deconvolution layer, an upsampling layer, and an activation function for image reconstruction. The deconvolution layer uses deconvolution (transposed convolution) to gradually map the feature map from low resolution back to high resolution, simultaneously extracting and enhancing feature information to provide data for subsequent image reconstruction or feature recovery. The upsampling layer employs upsampling algorithms (such as bilinear interpolation, nearest neighbor interpolation, or upsampling networks) to increase the size of the feature map, restore image details, and maintain feature consistency and coherence. The activation function is a non-linear activation function (such as ReLU, Leaky ReLU, PReLU, or ELU) to introduce non-linear characteristics into the decoding process, enhancing the model's expressive power and making the decoded image or features more realistic and natural. D 1( t ), D 21 ( t ), D 22 ( t )or D 3( t Using time-response images as input, models DCNN1 and DCNN2 are trained to extract information features from the time-response images. 21 DCNN 22 And DCNN3, which constructs a classification model by reconstructing errors.

[0051] In this embodiment, combined with Figure 4 and Figure 5 As shown, step 600: Measure the deflection response. Dm ( t The four components obtained by the above separation method D m 1(t), D m 21 ( t ), D m 22 ( t )and D m 3( t Input the trained DCNN1 and DCNN respectively. 21 DCNN 22 In DCNN3, the reconstruction error is used to classify the deflection response into two states: normal and abnormal. The criterion for a normal state is that all four classification modules produce normal results; the criterion for an abnormal state is...

[0052] (1) When DCNN1, DCNN 21 DCNN 22 Each of DCNN3 and DCNN3 having anomalies individually is considered anomaly state 1-4;

[0053] (2) When DCNN1, DCNN 21 DCNN 22 Anomalies in the combination of DCNN3 and DCNN3 are considered abnormal states (5-10).

[0054] (3) When DCNN1, DCNN 21 DCNN 22 Anomalies in the combination of DCNN3 and DCNN3 are considered abnormal states (11-14).

[0055] (4) When DCNN1, DCNN 21 DCNN 22 All four of them, including DCNN3, are considered abnormal, which is considered an abnormal state.15.

[0056] When an exception 15 occurs, it indicates that all modules are experiencing exceptions, requiring emergency intervention. These 15 exception states correspond to different exception types, as detailed in the table below.

[0057]

[0058] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementation methods. Clearly, many modifications and variations can be made based on the content of this specification. The selection and specific description of these embodiments in this specification are intended to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for detecting anomalies in bridge deflection monitoring response based on a deep convolutional autoencoder, characterized in that, Includes the following steps: Step 100: Collect bridge deflection monitoring response D over a time period t. t ); Step 200: Design a series of ideal filters and set the lower frequency cutoff limit according to the characteristics of the bridge. w l and frequency cutoff limit w u Ideal filters are based on w l and w u Components separated D 1( t ), D 2( t )and D 3( t ); Step 300: Select the window using the moving average method. D 2( t Separate into D 21 ( t )and D 22 ( t ); Step 400: Constructing the bridge deflection monitoring response under normal conditions D ( t Using the sample dataset, the deflection response components under normal conditions are obtained according to the separation method described above. D 1( t ), D 21 ( t ), D 22 ( t )and D 3( t ); Step 500: Design a deep convolutional network (DCNN) to... D 1( t ), D 21 ( t ), D 22 ( t )and D 3( t Using the time-series response images as input, the information feature extraction models DCNN1 and DCNN are trained. 21 DCNN 22 And DCNN3, which constructs a classification model by reconstructing errors, in which... D 1( t ), D 21 ( t ), D 22 ( t )and D 3( t This is obtained in step 400; Step 600: Measure the deflection response D m ( t The four components obtained according to the separation method above D m 1( t ), D m 21 ( t ), D m 22 ( t )and D m 3( t Input the trained DCNN1 and DCNN respectively. 21 DCNN 22 In DCNN3, the four components are divided into normal and abnormal states based on the reconstruction error. Specifically, step 100 involves collecting bridge deflection monitoring responses over a time period t. D ( t ), obtain spectral information through Fast Fourier Transform. F ( w ); In step 200, the ideal filter is based on... w l and w u Separate components D 1( t ), D 2( t )and D 3( t Specifically, the frequency w < w l The components are separated into [a specific composition] using a low-pass filter. F 1( w ), frequency w l ≤ w ≤ w u The components are separated into [a specific composition] using a bandpass filter. F 2( w ), frequency w > w u The components are separated into [a specific composition] using a high-pass filter. F 3( w ), by inverse Fourier transform, the components F 1( w ), F 2( w )and F 3( w They were converted into time domain signals respectively, and the results were obtained respectively. D 1( t ), D 2( t )and D 3( t ); The D 1( t )+ D 21 ( t The ) represents the low-frequency pseudo-static components caused by long-term downward deflection, annual temperature difference, monthly temperature difference, and daily temperature difference, and also represents the mid-frequency pseudo-static components caused by average wind field and traffic congestion. D 22 ( t ) represents the high-frequency pseudo-static component caused by sparse traffic, the D 3( t This represents the high-frequency dynamic components caused by traffic flow dynamics, environmental vibrations, and noise. The condition for determining the normal state in step 600 is that when all four classification modules have normal results, the deflection response is determined to be in a normal state. The abnormal state determination condition in step 600 is as follows: (1) When DCNN1, DCNN 21 DCNN 22 The existence of anomalies in DCNN1 and DCNN2, respectively, is considered as anomaly states 1-4. Anomaly state 1 indicates an anomaly in DCNN1, anomaly state 2 indicates an anomaly in DCNN2, and anomaly state 3 indicates an anomaly in DCNN2. 21 An anomaly has occurred; anomaly state 3 is for DCNN. 22 An anomaly has occurred; anomaly status 4 indicates that DCNN3 has encountered an anomaly. (2) When DCNN1, DCNN 21 DCNN 22 The combination of DCNN1 and DCNN3 exhibits anomalies, which are considered anomalous states 5-10. Anomalous state 5 refers to the combination of DCNN1 and DCNN3. 21 All showed abnormalities, with abnormal state 6 being DCNN1 and DCNN. 22 All showed anomalies; anomaly state 7 indicated that both DCNN1 and DCNN3 showed anomalies; anomaly state 8 indicated that DCNN... 21 and DCNN 22 All showed abnormalities, with abnormal state 9 being DCNN. 21 Both DCNN and DCNN3 exhibited anomalies, with anomaly state 10 being DCNN. 22 Both DCNN and DCNN3 showed anomalies; (3) When DCNN1, DCNN 21 DCNN 22 The combination of DCNN1, DCNN2, and DCNN3 exhibits anomalies, which are considered anomalous states 11-14. Anomalous state 11 refers to the combination of DCNN1, DCNN3, and DCNN4. 21 and DCNN 22 All showed abnormalities, with abnormal state 12 being DCNN1 and DCNN. 21 Both DCNN1 and DCNN3 exhibited anomalies, while anomaly state 13 was observed in DCNN1 and DCNN3. 22 Both DCNN and DCNN3 exhibited anomalies, with anomaly state 14 being DCNN. 21 DCNN 22 Both DCNN and DCNN3 showed anomalies; (4) When DCNN1, DCNN 21 DCNN 22 All four of them, including DCNN3, are considered abnormal, which is considered an abnormal state. The 15 abnormal states correspond to the following abnormal types: Abnormal state 1 corresponds to abnormal temperature difference; Abnormal state 2 corresponds to abnormal interference and abnormal noise; Abnormal state 3 corresponds to data loss and low resolution; Abnormal state 4 corresponds to abnormal re-run; Abnormal state 5 corresponds to abnormal temperature difference, abnormal interference, and abnormal noise; Abnormal state 6 corresponds to abnormal temperature difference, data loss, and low resolution; Abnormal state 7 corresponds to abnormal temperature difference and abnormal re-run; Abnormal state 8 corresponds to abnormal interference, abnormal noise, data loss, and low resolution; Abnormal state 9 corresponds to abnormal interference, abnormal noise, and abnormal re-run; Abnormal state 10 corresponds to data loss, low resolution, and abnormal re-run; Abnormal state 11 corresponds to abnormal temperature difference, abnormal interference, abnormal noise, data loss, and low resolution; Abnormal state 12 corresponds to abnormal temperature difference, abnormal interference, abnormal noise, and abnormal re-run; Abnormal state 13 corresponds to abnormal temperature difference, data loss, low resolution, and abnormal re-run; Abnormal state 14 corresponds to abnormal interference, abnormal noise, data loss, low resolution, and abnormal re-run; Abnormal state 15 corresponds to abnormal temperature difference, abnormal interference, abnormal noise, data loss, low resolution, and abnormal re-run.

2. The method for detecting anomalies in bridge deflection monitoring response based on a deep convolutional autoencoder according to claim 1, characterized in that, The deep convolutional network (DCNN) in step 500 includes an encoder module, a spatial-channel self-attention module, and a decoding module.

3. The method for detecting anomalies in bridge deflection monitoring response based on a deep convolutional autoencoder according to claim 2, characterized in that, The encoder module includes four convolutional sub-modules, each of which includes a convolutional layer, a downsampling layer, and an activation function.

4. The bridge deflection monitoring response anomaly detection method based on a deep convolutional autoencoder according to claim 3, characterized in that, The decoding module includes three deconvolution sub-modules, each of which includes a deconvolution layer, an upsampling layer, and an activation function.

Images