Bridge deformation monitoring method, computer device and computer readable storage medium

Abstract

To provide a bridge deformation monitoring method, a device and a storage medium for improving monitoring accuracy.SOLUTION: A method collects temperature data and deformation data of a set position on the main span and medium span cross section of a target bridge, classifies the temperature data and the deformation data using a data cleansing model, obtains the temperature data and the deformation data after the cleansing, which is obtained by training BiGRU-based neural network using a first training set, extracts temperature sensitive deformation data from the deformation data after the cleansing, compression treats the temperature sensitive deformation data and a temperature corresponding to the temperature sensitive deformation data, obtains the temperature data and the deformation data after the compression, constructs a second training set using the temperature data and the deformation data after the compression, trains a GRU network structure using the second training set, obtains a temperature sensitive deformation model, and monitors temperature sensitive deformation of a target bridge using the temperature sensitive deformation model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the technical field of bridge monitoring, and in particular to a bridge deformation monitoring method, a computer device, and a computer-readable storage medium. [Background technology]

[0002] The purpose of bridge safety monitoring is to install monitoring equipment and classify, organize, calculate, and analyze the collected time-series data to understand the operating status of bridges and ensure their safe operation. However, limited by the development of sensing and communication technologies, computer technologies, and data science technologies, bridge condition monitoring and safety warning theories and technologies based on intelligent processing and analysis of monitoring data are still in the exploratory stage. Furthermore, the complex environment in which high-speed rail bridges are located and the many factors involved often make it difficult to extract important information hidden in the monitoring data, making it difficult to solve the problem of bridge monitoring systems having a large amount of data but not accurate. Furthermore, the relatively high labor costs typically involved can lead to resource waste.

[0003] Vertical deformation is a very important monitoring item for bridge construction. The loads that cause vertical deformation of bridges are mainly temperature and vehicle loads. Therefore, the approach to establishing bridge linear reference value models can be divided into two categories: establishing a temperature-sensitive deflection model and establishing a vehicle-sensitive deflection model. Most previous studies have used linear regression methods to establish a temperature-sensitive deflection model between main girder temperature and main girder deflection. Generally, the main girder temperature is further extracted as the main girder top temperature, the main girder bottom temperature, and the main girder vertical temperature difference. These three temperature variables are then used to perform multi-dimensional linear regression analysis to obtain the corresponding temperature-sensitive deflection model. However, the accuracy of this approach still falls short of the actual needs of construction. Summary of the Invention [Problem to be solved by the invention]

[0004] An object of the present invention is to provide a bridge deformation monitoring method, device, and medium that improve monitoring accuracy. [Means for solving the problem]

[0005] To achieve the above object, the present invention provides the following solutions.

[0006] A bridge deformation monitoring method, comprising: collecting temperature data and deflection data at set positions on a main span and an intermediate span cross section of a target bridge; performing data classification of the temperature data and the deflection data using a data cleansing model, and obtaining cleansed temperature data and deflection data based on a data classification result, wherein the data cleansing model is obtained by training a BiGRU-based neural network using a first training set, and the data classification includes an outlier type and an inlier type; extracting temperature-sensitive deflection data from the cleansed deflection data; compressing the temperature-sensitive deflection data and the temperature corresponding to the temperature-sensitive deflection data to obtain compressed temperature data and deflection data; constructing a second training set using the temperature data and deflection data after compression, and training a GRU network structure using the second training set to obtain a temperature-sensitive deflection model; and monitoring the temperature-sensitive deformation of the target bridge using the temperature-sensitive deflection model.

[0007] Optionally, the set positions include a main spar diagonal and a main spar.

[0008] Optionally, the BiGRU-based neural network includes a sequence input layer, a first BiGRU, a second BiGRU, a first fully-connected layer, a second fully-connected layer, a Softmax layer, and a classification output layer, which are connected in sequence; The sample data in the first training set includes first input data and first label data, the first input data being feature data obtained based on a time series of temperature data or feature data obtained based on a time series of deflection data, the first label data being a classification corresponding to the first input data, and the classification including an abnormal type different from a normal type.

[0009] Optionally, extracting the temperature sensitive deflection data from the cleansed deflection data specifically includes: Extracting temperature sensitive deflection data from the cleansed deflection data using singular spectrum analysis methods.

[0010] Optionally, compressing the temperature-sensitive deflection data and the temperature corresponding to the temperature-sensitive deflection data to obtain compressed temperature data and deflection data can be specifically performed by: The temperature-sensitive deflection data and the temperature data corresponding to the temperature-sensitive deflection data are compressed using a method of taking an average value for each set time period, and the temperature data and deflection data after compression are obtained.

[0011] Optionally, the sample data in the second training set includes second input data and second label data, the second input data including a main spar top temperature, a main spar bottom temperature, and a main spar vertical temperature difference, and the second label data is temperature-sensitive deflection data corresponding to the second input data.

[0012] Optionally, training a GRU network structure using the second training set to obtain a temperature-sensitive deflection model specifically includes: The method includes using a t-test method to test the trained GRU network structure, and if it fails, continuing to train the GRU network structure until it passes the test, thereby obtaining the temperature-sensitive deflection model.

[0013] Optionally, monitoring the temperature-sensitive deformation of the target bridge using the temperature-sensitive deflection model specifically includes: inputting temperature data to be detected of the target bridge into the temperature-sensitive deflection model to obtain a deflection prediction value; and issuing an alarm if the deflection prediction exceeds an alarm threshold.

[0014] The computer device includes a memory, a processor, and a computer program stored on the memory and operable on the processor, the processor executing the computer program to implement the steps of the bridge deformation monitoring method.

[0015] A computer-readable storage medium has a computer program stored thereon, which, when executed by a processor, implements the steps of the bridge deformation monitoring method. [Effects of the Invention]

[0016] According to the specific embodiments of the present invention, the present invention discloses the following technical effects.

[0017] The present invention adopts a temperature-sensitive deflection model based on a GRU neural network, which requires less computing resources, can reduce the risk of overfitting, and has a fast training speed. It also uses deep learning technology to mine the temperature and temperature-sensitive deflection data features in the monitoring data to construct a second training set, and obtains a temperature-sensitive deflection model by training the GRU network structure, thereby improving the monitoring accuracy. [Brief explanation of the drawings]

[0018] [Figure 1] 1 is a flowchart of a bridge deformation monitoring method according to a first embodiment of the present invention. [Figure 2] 1 is a specific flowchart of a bridge deformation monitoring method according to a first embodiment of the present invention. [Figure 3] 1 is a schematic diagram of a GRU structure according to a first embodiment of the present invention; [Figure 4]FIG. 1 is a schematic diagram of a data cleansing model structure according to a first embodiment of the present invention. [Figure 5] 1 is a flowchart of extracting temperature-sensitive deflection using an SSA method according to a first embodiment of the present invention; [Figure 6] 1 is a schematic diagram of a temperature-sensitive deflection model structure according to a first embodiment of the present invention; [Figure 7] 1 is a warning flowchart according to the first embodiment of the present invention. [Figure 8] FIG. 1 is a diagram illustrating the internal structure of a computer device. DETAILED DESCRIPTION OF THE INVENTION

[0019] An object of the present invention is to provide a bridge deformation monitoring method, device, and medium that improve monitoring accuracy.

[0020] The gate-controlled cycle unit is a variant of a cycle neural network. Compared with other variants of long short-term memory networks, the GRU neural network has the following advantages: while the LSTM has more training parameters, increasing complexity and the risk of overfitting, the GRU has fewer parameters, is relatively more computationally efficient, and has a faster training speed; the GRU only includes an update gate and a reset gate, while the LSTM has an input gate, a forget gate, and an output gate, making the GRU easier to train when there is relatively little data or relatively large data noise; and because it has fewer parameters, the GRU has an advantage in some resource-limited environments. Overall, the GRU is a lightweight version of the LSTM, and the core technology of the present invention is the GRU neural network.

[0021] Example 1 As shown in FIGS. 1 and 2, the bridge deformation monitoring method in this embodiment includes the following steps:

[0022] Step 101: Collect temperature data and deflection data at set positions on the main span and mid-span cross section of the target bridge.

[0023] Step 102: Perform data classification of the temperature data and the deflection data using a data cleansing model, and obtain cleansed temperature data and deflection data based on the data classification result, where the data cleansing model is obtained by training a BiGRU-based neural network using a first training set, and the data classification includes an outlier type and a normal value type.

[0024] Step 103: Extract temperature-sensitive deflection data from the cleansed deflection data.

[0025] Step 104: The temperature-sensitive deflection data and the temperature corresponding to the temperature-sensitive deflection data are compressed to obtain compressed temperature data and deflection data.

[0026] Step 105: Construct a second training set using the temperature data and deflection data after compression, and train a GRU network structure using the second training set to obtain a temperature-sensitive deflection model.

[0027] Step 106: Monitor the temperature-sensitive deformation of the target bridge using the temperature-sensitive deflection model.

[0028] Here, in step 101, the set positions include the main girder diagonal and the main girder, and specifically, temperature and deflection data are collected from the main girder diagonal and the main girder at the main span and mid-span cross section, respectively.

[0029] The main girder diagonal includes the top and bottom plates of the upper and lower chords, and the main girder includes the center of the lower chord of the main girder.

[0030] Step 101 specifically includes selecting a main girder diagonal and a main girder in the mid-span section of the monitored bridge, and respectively setting a first Internet of Things temperature sensor W1, a second Internet of Things temperature sensor W2, a third Internet of Things temperature sensor W3, a fourth Internet of Things temperature sensor W4, and an Internet of Things deflection sensor N. Generally, the temperature data correlation at the mid-span position in the mid-span section is particularly obvious, and the data is relatively smooth, so points at the mid-span position are selected to collect data.

[0031] Here, step 102 identifies and cleanses the temperature and deflection data using a BiGRU-based neural network. Obtaining cleansed temperature and deflection data based on the data classification result specifically includes excluding outlier-type data in the data classification and obtaining cleansed temperature and deflection data.

[0032] The BiGRU neural network is a bidirectional propagation neural network established to overcome the problem that the GRU neural network can only capture previous values ​​and cannot use future values. The BiGRU combines two backward single hidden GRU layers into the same output. In this way, the sub-output layer can use previous and future values.

[0033] The internal structure of the GRU is shown in Figure 3.

[0034] The GRU neural network is constructed by Equation (1) to Equation (4), JPEG2025155487000002.jpg59170where h t represents the system state at time t, and h t-1 represents the system state at time t-1, and z t represents the update gate at time t, and r t represents the reset gate at time t, and x t represents the input at time t, σ represents the sigmoid function, and w zrepresents the update gate weight, and w r represents the reset gate weight, and w is the candidate system state at time t. JPEG2025155487000003.jpg5170 represents the power.

[0035] Here, the bidirectional propagation process is constructed by Equations (5) to (7), JPEG2025155487000004.jpg42170JPEG2025155487000005.jpg6170Represents the system status.

[0036] Before the temperature data and the deflection data are input into the data cleansing model, the temperature data and the deflection data are respectively organized into a sequence format to ensure data alignment and proper standardization.

[0037] As shown in FIG. 4, the BiGRU-based neural network includes a sequence input layer, a first BiGRU, a second BiGRU, a first fully connected layer, a second fully connected layer, a Softmax layer, and a classification output layer, which are connected in sequence.

[0038] The sample data in the first training set includes first input data and first label data, the first input data being feature data (nine feature representations) obtained based on a time series of temperature data or feature data (nine feature representations) obtained based on a time series of deflection data, the first label data being a classification corresponding to the first input data, and the classification including an abnormal type different from a normal type.

[0039] The feature data includes nine representations, each of which is a vector. Feature extraction is performed every minute, and the training sample size for one hour is 9 × 60, depending on the number of representations selected.

[0040] The nine selectable feature representations are: 1) standard deviation of the data, 2) difference between the maximum and minimum values ​​of the data, 3) mean value divided by data range, 4) mean value of the data, 5) standard deviation of the data, 6) median value of the data, 7) mean square level of the data, 8) maximum value divided by 80th percentile value, and 9) mean value divided by 80th percentile value.

[0041] For the collected temperature and deflection raw data signals, the mark data set is classified according to the sensor anomaly type (outlier type) and the distribution characteristics of different anomaly types are obtained.

[0042] Mark outlier types include data gaps, data jumps, data drifts and data trend anomalies.

[0043] Here, when setting the algorithm threshold, the false negative rate of the data is reduced to 0.5% or less as much as possible, and false positives are appropriately mitigated.

[0044] The feature data formation dataset extracted in step 102 is input as a matrix, and the sample size is represented as n×9×60.

[0045] The number of neurons in the hidden layer in the BiGRU-based neural network is set to 16, SELU is used as the activation function in the hidden layer, and Softmax is used in the output layer.

[0046] The SELU is a scaled version of the exponential linear unit, and if properly initialized, feedforward networks using linear layers are self-normalizing.

[0047] The output of each layer has a mean value approximately equal to 0 and a standard deviation equal to 1, which helps prevent the vanishing or exploding gradient problem and allows the construction of deep networks.

[0048] The loss function adopted for training BiGRU-based neural networks is the Huber Loss function. Huber Loss can balance the sensitivity to small and large errors. The formula for the Huber Loss function is as follows:

[0049] JPEG2025155487000006.jpg20170JPEG2025155487000007.jpg8170 is the predicted data, x is the model input data, δ is a parameter, and the magnitude of the δ value determines the weighting of the Huber Loss mean squared error and mean absolute error.

[0050] The BiGRU model has a relatively good cleansing effect on different data anomaly types, and the final output temperature and deflection data are used to establish the subsequent model.

[0051] Here, in step 103, as shown in FIG. 5, extracting the temperature-sensitive deflection data from the cleansed deflection data using the singular spectrum analysis method specifically includes: The bridge monitoring method includes continuously monitoring the vehicle-induced deflection of the bridge by installing strain gauges, displacement sensors, and accelerometers at the center of the lower chord of the main girder that requires monitoring, where W3 and W4 record the temperature changes of the main girder to separate the temperature effects from the data. The Internet of Things sensors are used to complete the use of the SSA method.

[0052] The data from all sensors is time-synchronized and the data is primarily analyzed during peak traffic hours.

[0053] Based on statistical probability knowledge, the data is processed in an initial stage, and the strain data is analyzed to identify the types of vehicles (passenger cars, freight cars, trains, etc.) crossing the bridge at different times, and the loads are estimated based on the vehicle types.

[0054] Based on the monitoring results, the time series data is divided into individual vehicle crossing events, and the data for each event is analyzed separately to extract the vehicle-responsive deflection.

[0055] The time series of vehicle-sensitive deflection data is constructed into a trajectory matrix formed by selecting an appropriate delay window size and overlap window. Singular value decomposition (SVD) is performed on the constructed trajectory matrix.

[0056] The delay window size L is smaller than half the length of the selected data sequence. Starting from the beginning of the time series, take a data segment of length L as the first column of the trajectory matrix, then move one time unit to the right and take the next data segment of length L as the second column. This process is repeated until the end of the time series. The dimension of the final trajectory matrix is ​​L × K (K is the total number of columns, which is equal to the time series length minus L plus 1). Depending on different actual situations, the specific value of L needs to be continuously tested to obtain optimal results.

[0057] The singular vectors selected from the singular value decomposition result are used to reconstruct the original data by inverse transformation, and the reconstructed time series contains the main vehicle-sensitive deflection signal, which generally corresponds to a medium singular value.

[0058] Based on the reconstructed time series, the cleansed deflection data is subjected to mode matching or sliding window techniques to filter out the vehicle-sensitive deflection data, and the final temperature-sensitive deflection data is obtained.

[0059] Here, step 104 specifically includes the following steps: The temperature-sensitive deflection data and the temperature data corresponding to the temperature-sensitive deflection data are compressed using a method of taking an average value for each set time period, and the compressed temperature data and deflection data are obtained.

[0060] The sample data in the second training set includes second input data and second label data, and the second input data is a main spar top temperature W Tand the temperature at the bottom of the main girder W B and the temperature difference W in the vertical direction of the main girder C and the second label data is temperature-sensitive deflection data corresponding to the second input data.

[0061] JPEG2025155487000008.jpg43170, where W1 represents the temperature collected by the first Internet of Things temperature sensor, W2 represents the temperature collected by the second Internet of Things temperature sensor, W3 represents the temperature collected by the third Internet of Things temperature sensor, and W4 represents the temperature collected by the fourth Internet of Things temperature sensor.

[0062] The temperature data and temperature-sensitive deflection data for the first five hours of the current time are acquired, i.e., the first 30 data points are input, and enumerated from the 30th time point onwards, i.e., data points for every 30 time points are input as a sequence. For example, data sets 1 to 30 are input at the 30th time point, and data sets 2 to 31 are input at the 31st time point, and inference is made based on this. As shown in Table 1.

[0063] The first 80% of these data are used as the training set and the last 20% as the test set.

[0064] As shown in Figure 6, the GRU network structure has an input layer with a total of 30 units, and the dimension of the data input is 3, i.e., W T W B W C The hidden layer has a total of 64 units, X1, X2, ..., X in Figure 6. t are all inputs to the GRU network structure, and GRU1, GRU2, ..., GRU t are all GRU, H1, H2, ..., and H t-1 are the outputs of each GRU, and FC1, FC2, ..., FC t are all fully connected layers, and Y is the output of the GRU network structure. The input / output modes are as shown in Table 1.

[0065] JPEG2025155487000009.jpg14170JPEG2025155487000010.jpg65170

[0066] The GRU network structure uses SELU (Scaled ELU) as the activation function and Huber Loss as the loss function.

[0067] In the output layer, the multi-dimensional data is normalized to one deflection output data, namely the neural network regression value of the temperature-sensitive deflection.

[0068] In the backpropagation process, the Adam algorithm is used to continuously optimize the GRU parameters. After a certain number of forward calculations and backpropagation processes, a temperature-sensitive deflection model based on the trained GRU is obtained.

[0069] During training, the test set is used to monitor the model's performance and avoid overfitting.

[0070] Here, step 105 specifically includes the following steps: The trained GRU network structure is tested using a t-test (student's test), and if it fails, the GRU network structure is continuously trained until it passes the test, thereby obtaining the temperature-sensitive deflection model.

[0071] Furthermore, by performing a t-test between the GRU neural network regression value of the temperature-sensitive deflection and the actual measured value, and by reducing and testing the deflection deviation, it is possible to identify relatively small deflection abnormalities.

[0072] The t-test formula is expressed as follows:

[0073] JPEG2025155487000011.jpg103170

[0074] JPEG2025155487000012.jpg6170

[0075] If the calculated t-value exceeds the threshold in the t-distribution table for the given degrees of freedom and significance level, the model is considered unsuccessful in training and needs to be retrained.

[0076] Here, as shown in FIG. 7, step 106 specifically includes the following steps: inputting temperature data to be detected of the target bridge into the temperature-sensitive deflection model to obtain a deflection prediction value; and issuing an alarm if the deflection prediction exceeds an alarm threshold.

[0077] An alarm threshold is set based on the modeling results, and an alarm is issued if the predicted value exceeds the threshold.

[0078] According to the actual conditions of the bridge, a threshold value for deflection change is set in combination with the theoretical deflection value output from the temperature-sensitive deflection model, and different conditions are simulated to test the effectiveness of the threshold value.

[0079] The predicted deflection value, the degree to which the threshold is exceeded, and recommended countermeasures (pre-settings) are sent to the maintenance personnel and bridge managers via an automated system in the form of email, message, or dedicated application program.

[0080] Furthermore, the specific location where the abnormality occurred is provided, and a specific alarm level is determined based on the level of the deflection abnormality. A graph comparing the predicted deflection data with the historical data is provided, highlighting the abnormality range.

[0081] Furthermore, depending on the degree to which the predicted deflection exceeds the threshold, it is classified into "normal monitoring," "warning," "primary alarm," and "secondary alarm."

[0082] Furthermore, 0% to 5% is normal monitoring, 5% to 10% is a warning, 10% to 20% is a primary alarm, and 20% or more is a secondary alarm.

[0083] Trigger emergency response flows, such as temporarily closing bridges, conducting on-site inspections, and initiating maintenance work.

[0084] The beneficial effects of the solution of the present invention are as follows:

[0085] This breaks the current lack of artificial intelligence management in the field of temperature-sensitive deformation of bridges. Compared to traditional LSTM prediction methods, GRUs require fewer computing resources, reduce the risk of overfitting, and have a faster training speed. While the data sets used in this invention are generally relatively short in time span, GRUs can capture time dependencies more effectively than LSTMs in relatively short sequences. This invention uses deep learning techniques to mine temperature and temperature-sensitive deformation data features from monitoring data, and combines this with mathematical and statistical knowledge to achieve accurate predictions once training is complete.

[0086] After the alarm system is established, manual intervention can be greatly reduced and automated bridge detection and alarming can be realized. The adopted Internet of Things sensors can realize wireless data transmission, which not only reduces the amount of wiring work in the early stages but also reduces line maintenance costs in the later stages.

[0087] <Example 2> A computer-readable storage medium has a computer program stored thereon, which, when executed by a processor, implements the steps of the bridge deformation monitoring method in the first embodiment.

[0088] Example 3 A computer program product includes a computer program, which, when executed by a processor, implements the steps of the bridge deformation monitoring method in the first embodiment.

[0089] Example 4 The internal structure of the computer device may be as shown in FIG. 8. The computer device includes a processor, a memory, an input / output interface (abbreviated as Input / Output, I / O), and a communication interface. The processor, memory, and I / O interface are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. An operating system, a computer program, and a database are stored in the non-volatile storage medium. The internal memory provides an environment for the operating system and the computer program to run in the non-volatile storage medium. The database of the computer device is used to store tasks to be processed. The I / O interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, the bridge deformation monitoring method of Example 1 is realized.

[0090] It should be clarified that the subject information (including but not limited to subject device information, subject personal information, etc.) and data (including but not limited to data for analysis, data for storage, data for presentation, etc.) involved in the present invention are all information and data with the subject's permission or sufficient permission of each party.

[0091] As will be understood by those skilled in the art, realizing all or part of the steps in the methods of the above embodiments can be completed by instructing relevant hardware via a computer program. The computer program may be stored in a non-volatile computer-readable storage medium, and when executed, may include the steps of the above-described method embodiments. Any reference to a memory, database, or other medium used in the embodiments of the present invention may include at least one of a non-volatile memory and a volatile memory. Non-volatile memory may include read-only memory, magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive memory, magnetic change memory, ferroelectric memory, phase-change memory, graphene memory, etc. Volatile memory may include random access memory or external high-speed cache memory, etc. The database used in the embodiments of the present invention may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in various embodiments of the present invention may be, but are not limited to, general purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, and the like.

Claims

1. A bridge deformation monitoring method, comprising: collecting temperature data and deflection data at set positions on a main span and an intermediate span cross section of a target bridge; performing data classification of the temperature data and the deflection data using a data cleansing model, and obtaining cleansed temperature data and deflection data based on a data classification result, wherein the data cleansing model is obtained by training a BiGRU-based neural network using a first training set, and the data classification includes an outlier type and an inlier type; extracting temperature-sensitive deflection data from the cleansed deflection data; compressing the temperature-sensitive deflection data and the temperature corresponding to the temperature-sensitive deflection data to obtain compressed temperature data and deflection data; constructing a second training set using the temperature data and deflection data after compression, and training a GRU network structure using the second training set to obtain a temperature-sensitive deflection model; The GRU network is constructed by the following equations (1) to (4): Here, h ｔ represents the system state at time t, and h ｔ－１ represents the system state at time t-1, and z ｔ represents the update gate at time t, and r ｔ represents the reset gate at time t, and x ｔ represents the input at time t, σ represents the sigmoid function, and w ｚ represents the update gate weight, and w ｒ represents the reset gate weight, and w is the candidate system state at time t. Representing power, Here, the bidirectional propagation process is constructed by equations (5) to (7): Representing the system state, and organizing the temperature data and the deflection data into a sequence format to ensure data alignment and proper standardization before inputting the temperature data and the deflection data into the data cleansing model, and monitoring the temperature-sensitive deformation of the target bridge using the temperature-sensitive deflection model.

2. The bridge deformation monitoring method according to claim 1 , wherein the set positions include a main girder diagonal and a main girder.

3. The BiGRU-based neural network includes a sequentially connected sequence input layer, a first BiGRU, a second BiGRU, a first fully-connected layer, a second fully-connected layer, a Softmax layer, and a classification output layer; 2. The bridge deformation monitoring method according to claim 1, wherein the sample data in the first training set includes first input data and first label data, the first input data being feature data obtained based on a time series of temperature data or feature data obtained based on a time series of deflection data, the first label data being a classification corresponding to the first input data, and the classification including an abnormal type different from a normal type.

4. Specifically, extracting the temperature-sensitive deflection data from the cleansed deflection data includes:

2. The method of claim 1, further comprising extracting temperature sensitive deflection data from the cleansed deflection data using a singular spectrum analysis method.

5. Specifically, compressing the temperature-sensitive deflection data and the temperature corresponding to the temperature-sensitive deflection data to obtain the compressed temperature data and deflection data includes the following steps: The bridge deformation monitoring method according to claim 1, characterized in that it includes compressing the temperature-sensitive deflection data and the temperature data corresponding to the temperature-sensitive deflection data using a method of taking an average value for each set time period, and obtaining the compressed temperature data and deflection data.

6. The bridge deformation monitoring method of claim 1, wherein the sample data in the second training set includes second input data and second label data, the second input data including the main girder top temperature, the main girder bottom temperature, and the main girder vertical temperature difference, and the second label data is temperature-sensitive deflection data corresponding to the second input data.

7. The second training set is used to train a GRU network structure to obtain a temperature-sensitive deflection model, specifically, 2. The bridge deformation monitoring method according to claim 1, further comprising: testing the trained GRU network structure using a t-test method; if the test fails, continuing to train the GRU network structure until the test passes, thereby obtaining the temperature-sensitive deflection model.

8. Specifically, monitoring the temperature-sensitive deformation of the target bridge using the temperature-sensitive deflection model includes: inputting temperature data to be detected of the target bridge into the temperature-sensitive deflection model to obtain a deflection prediction value; 2. The bridge deformation monitoring method according to claim 1, further comprising issuing an alarm if the predicted deflection value exceeds an alarm threshold.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and operable to run on the processor, the processor executing the computer program to implement the steps of the method of any one of claims 1 to 8.

10. A computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 8.

Images