Adaptive monitoring method, device and medium for stress in subway tunnel walls

Abstract

This invention discloses an adaptive monitoring method, device, and medium for stress in subway tunnel walls. The method includes the following steps: acquiring vibration signals collected by train vibration sensors and optical signals collected in real time by distributed fiber optic sensors; demodulating and calculating the actual stress under thermo-mechanical coupling; extracting the dynamic displacement of the tunnel wall from the vibration signals to calculate the dynamic stress; and further calculating the actual stress under thermo-mechanical-vibration multi-field coupling. The real-time monitored stress signals are denoised using wavelet transform; Kalman filtering is used to predict the corrected stress data based on the denoised stress signals and the actual stress data under multi-field coupling; and the corrected stress data is input into an LSTM-SVM model to obtain the final stress prediction result. This invention can comprehensively and continuously monitor the stress state of subway tunnel walls, while improving the monitoring accuracy, stability, and robustness in complex and dynamic tunnel environments.

Description

Technical Field

[0001] This invention relates to the field of stress monitoring technology, and in particular to an adaptive monitoring method, device and medium for stress monitoring of subway tunnel walls. Background Technology

[0002] Traditional tunnel stress monitoring typically uses point-to-point sensors such as stress sensors for data acquisition. These sensors, including resistance strain gauges and piezoelectric sensors, can provide stress data within a local area. However, the monitoring range and spatial distribution are limited by the number and location of the sensors, making it impossible to comprehensively monitor stress changes throughout the entire tunnel. Consequently, they cannot fully reflect the health status of the tunnel structure, especially in long tunnels or under complex geological conditions, where there are many blind spots. Furthermore, these sensors have long monitoring cycles, making it difficult to meet the needs of real-time monitoring. They are also often sensitive to environmental factors (such as temperature and electromagnetic interference), which can easily lead to measurement errors. Therefore, they suffer from problems such as poor real-time performance, narrow coverage, inability to conduct continuous monitoring, and low monitoring accuracy.

[0003] Fiber optic sensors utilize the optical properties of optical fibers to sense physical quantities such as temperature, pressure, and vibration, enabling comprehensive and continuous capture of changing signals. However, in current technology, fiber optic sensors are typically used for tunnel deformation monitoring to track tunnel deformation. Furthermore, directly using fiber optic sensors for tunnel wall stress monitoring only detects mechanical stress. In actual subway tunnel operation, the stress on the tunnel walls is affected by the coupling effects of multiple physical fields, such as train vibration and temperature changes. Therefore, the mechanical stress detected directly by fiber optic sensors will deviate from the actual stress value, making accurate stress monitoring difficult.

[0004] During subway operation, it is often necessary to monitor the stress distribution in subway tunnel walls to identify stress concentration areas and potential weak points in advance, thereby guiding tunnel design optimization. However, sensors only collect real-time stress data. Existing technologies for predicting stress distribution typically use a large amount of stress measurement data to train neural network models. However, tunnel stress changes are affected by various complex factors, exhibiting nonlinearity, time dependence, and uncertainty. Traditional neural network models often struggle to handle high-noise, complex time-series data, resulting in unstable prediction results and low accuracy in complex environments. Summary of the Invention

[0005] The technical problem to be solved by this invention is: in view of the technical problems existing in the prior art, this invention provides a simple implementation method, device and medium for adaptive monitoring of subway tunnel wall stress, which has high monitoring accuracy and stability, strong environmental adaptability and robustness. It can comprehensively and continuously monitor the stress state of subway tunnel walls, while improving the monitoring accuracy, stability and robustness in complex and dynamic tunnel environments.

[0006] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows:

[0007] An adaptive monitoring method for stress in subway tunnel walls, comprising the following steps:

[0008] The system monitors vibration signals collected by train vibration sensors and optical signals collected in real time by distributed fiber optic sensors installed on the subway tunnel wall. After demodulating the monitored optical signals, stress signals are obtained, and temperature changes and Bragg wavelength changes are extracted. The actual stress under thermo-mechanical coupling is calculated based on the extracted temperature changes and Bragg wavelength changes. The dynamic displacement of the tunnel wall caused by train vibration is extracted based on the monitored vibration signals, and dynamic stress is calculated based on the dynamic displacement of the tunnel wall. The actual stress data under thermo-mechanical-vibration multi-field coupling is calculated based on the actual stress under thermo-mechanical coupling and the dynamic stress.

[0009] The stress signal monitored in real time by the distributed optical fiber sensor is denoised using wavelet transform. The denoised stress signal is used as the observation data, and the actual stress data under the thermo-mechanical-vibration multi-field coupling is used as the initial stress state data. Kalman filtering is used for prediction to obtain the corrected thermo-mechanical-vibration coupled stress data.

[0010] The corrected thermo-mechanical-vibration coupled stress data are input into a pre-trained stress prediction model to obtain the final stress prediction result of the subway tunnel wall. The stress prediction model is obtained by training an LSTM-SVM model using a training dataset composed of historical stress data. The LSTM (Long Short-Term Memory)-SVM (Support Vector Machine) model uses the LSTM model to obtain the initial stress prediction result, and uses the output of the LSTM model as the input of the SVR model. The SVR model corrects the output of the LSTM model to obtain the final stress prediction result.

[0011] Furthermore, the step of calculating the actual stress data under the thermo-mechanical-vibration multi-field coupling effect based on the actual stress under the thermo-mechanical coupling effect and the dynamic stress includes:

[0012] The dynamic displacement of the tunnel wall is extracted from the collected vibration signals, and the real-time temperature change and Bragg wavelength change are extracted from the demodulated stress signals.

[0013] The actual stress under thermo-mechanical coupling is calculated based on the change in Bragg wavelength and the real-time temperature change.

[0014] Dynamic strain is calculated based on the real-time dynamic displacement of the tunnel wall, and then the dynamic stress caused by train vibration is calculated based on the calculated dynamic strain.

[0015] The actual stress under the thermo-mechanical coupling effect and the dynamic stress caused by the train vibration are superimposed to obtain the actual stress under the thermo-mechanical-vibration multi-field coupling effect.

[0016] Furthermore, the calculation expression for the actual stress under the aforementioned thermo-mechanical coupling effect is as follows:

[0017] ,

[0018] in, This represents the actual stress under thermo-mechanical coupling. Indicates the effective elastic coefficient. Indicates the effective refractive index, This represents the elastic modulus of the optical fiber. This represents the total change in the Bragg wavelength. Indicates the Bragg wavelength. This represents the coefficient of thermal expansion of the optical fiber material. Indicates the thermo-optic coefficient. It represents the amount of temperature change.

[0019] Furthermore, the calculation expression for the dynamic stress caused by train vibration is as follows:

[0020] ,

[0021] in, This indicates the dynamic stress caused by train vibration. This represents the elastic modulus of the optical fiber. Indicates dynamic strain;

[0022] Actual stress under thermo-mechanical-vibration multi-field coupling The calculation expression is:

[0023] = + ,

[0024] When train vibration causes elliptic deformation of the tunnel wall, the dynamic stress caused by the train vibration Mean tangential stress :

[0025] ,

[0026] in, Where is the tunnel radius. For train vibration amplitude, For frequency, For the material's Poisson's ratio, For time.

[0027] Furthermore, the noise reduction processing of the stress data monitored in real time by the distributed fiber optic sensor using wavelet transform includes:

[0028] Based on the frequency characteristics and noise distribution of the stress data, a wavelet basis is selected. The selected wavelet basis is used to perform multi-level wavelet decomposition on the stress signal monitored in real time by the distributed optical fiber sensor. In each decomposition, the low-frequency coefficients are obtained by convolving the input stress signal with the scaling function and convolving the input stress signal with the wavelet basis function. The next level further decomposes the low-frequency coefficients obtained from the previous level. In each decomposition, the high-frequency coefficients are denoised using a preset dynamic threshold to remove high-frequency noise. The dynamic threshold is dynamically set according to the parameters of the high-frequency coefficients in each level. Finally, the low-frequency coefficients obtained after multi-level wavelet decomposition and the denoised high-frequency coefficients are subjected to inverse wavelet transform to obtain the denoised stress data.

[0029] The stress data after the first denoising is processed by an adaptive filter to remove low-frequency noise, and the filter weights are continuously adjusted to adaptively track the changes in low-frequency noise, resulting in stress data after the second denoising.

[0030] Furthermore, the step of using a preset dynamic threshold to denoise the high-frequency coefficients in each layer of decomposition to remove high-frequency noise includes:

[0031] If the absolute value of the coefficient corresponding to an effective signal exceeding a specified proportion in the current high-frequency coefficients is greater than the dynamic threshold, or the absolute value of the coefficient corresponding to noise exceeding a specified proportion is significantly less than the dynamic threshold, then a hard thresholding method is used based on the dynamic threshold; otherwise, a soft thresholding method is used based on the dynamic threshold. If a hard thresholding method is used... hour, , hour, Indicates the first j High-frequency coefficients after layer wavelet decomposition This represents the high-frequency coefficients after noise reduction processing;

[0032] If a soft thresholding method is used for processing... hour, , At that time, the high-frequency coefficients of the current layer will be... The high-frequency coefficients after being reduced in size are used as the noise reduction coefficients. ;

[0033] Dynamic threshold The calculation expression is:

[0034] ,

[0035] in, For the first Standard deviation of the high frequency coefficient of the layer For the first The length of the high-frequency coefficient of the layer, This refers to the factors affecting tunnel vibration and noise.

[0036] Furthermore, using the denoised stress signal as observation data and the actual stress data under the thermo-mechanical-vibration multi-field coupling as the initial stress state data, extended Kalman filtering is used for prediction, including:

[0037] according to Perform state prediction, according to To predict the error covariance, according to The Kalman gain was calculated. ,according to Perform a status update, according to Perform error covariance update, where, The Kalman gain at time k is... Let $\mathbf{k}$ be the covariance of the stress state after the update at time $k$. Let $\mathbf{k}$ be the covariance of the predicted stress state at time $k$. For the observation matrix, The value at time k is the observed value, which is the stress signal obtained after wavelet transform and noise reduction from the stress signal monitored in real time by the distributed fiber optic sensor at time k. To observe the noise covariance, To predict the stress state at time k, the initial time is... This refers to the actual stress data under the combined effects of thermo-mechanical-vibration multi-field coupling. It is the identity matrix. The updated stress state estimate at time k is the corrected thermo-mechanical-vibrational coupled stress data. and These are Jacobian matrices, respectively.

[0038] Furthermore, the step of inputting the corrected thermo-mechanical-vibration coupled stress data into a pre-trained stress prediction model to obtain the stress prediction result of the subway tunnel wall includes:

[0039] The corrected thermo-mechanical-vibration coupled stress data is input into an LSTM model that has been trained using historical stress data. The LSTM model then predicts the stress changes at future moments based on the historical stress data to obtain the stress prediction result.

[0040] The output of the LSTM model is used as the input feature of the SVR model. The SVR model performs nonlinear optimization on the stress prediction value output by the LSTM model to obtain the optimized prediction result.

[0041] The outputs of the LSTM model and the SVR model are fused to obtain the fused prediction result;

[0042] The final stress prediction result is obtained based on the optimized prediction result and the fused prediction result.

[0043] Furthermore, it also includes early warning of multi-field coupled stress exceeding limits according to the following decision rules:

[0044]

[0045] in, This indicates the judgment indicator; 1 indicates that an over-limit warning has been triggered, and 0 indicates that an over-limit warning has not been triggered. These are the stress prediction values ​​output by the LSTM-SVR model. As the baseline threshold, For tunnel environment sensitivity coefficient, The standard deviation of historical stress.

[0046] An adaptive monitoring device for stress in subway tunnel walls includes a processor and a memory, wherein the memory stores a computer program and the processor executes the computer program to perform the method described above.

[0047] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0048] Compared with existing technologies, the advantages of this invention are as follows: This invention utilizes distributed optical fiber sensors installed on the tunnel wall to monitor the stress signal of the tunnel wall in real time. After demodulating the optical signal, it converts it into stress data considering the multi-field coupling of thermo-mechanical-vibration. The actual stress is obtained by comprehensively considering the actual stress under thermo-mechanical coupling and the dynamic stress caused by train vibration. This can accurately convert the mechanical stress collected by the optical fiber sensor into the actual stress of the subway tunnel wall under the multi-physics coupling. The calculated actual stress under the multi-physics coupling is used as the initial state data, and the stress signal collected by the distributed optical fiber sensor is used as the observation data after noise reduction. Kalman filtering is used to predict and obtain the corrected thermo-mechanical-vibration coupling stress data, which can effectively correct errors and obtain more accurate multi-physics coupling stress data. Then, a stress prediction model constructed using the LSTM-SVM model is used for stress prediction. This can combine the advantages of the LSTM model and the SVM model in time dependence and nonlinear optimization, further improving the prediction accuracy, reliability and robustness, and meeting the real-time and accuracy requirements of subway tunnel stress monitoring. Attached Figure Description

[0049] Figure 1 This is a schematic diagram illustrating the implementation process of the adaptive monitoring method for subway tunnel wall stress in this embodiment.

[0050] Figure 2 This is a schematic diagram illustrating the process of extracting optical signals and calculating stress values ​​based on fiber optic sensing technology in this embodiment.

[0051] Figure 3 This is a schematic diagram illustrating the implementation process of wavelet transform in this embodiment.

[0052] Figure 4 This is a schematic diagram of the prediction and update implementation process of Kalman filtering in this embodiment.

[0053] Figure 5 This is a schematic diagram illustrating the training and prediction process of the stress prediction model in this embodiment. Detailed Implementation

[0054] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0055] like Figure 1 As shown, the steps of the adaptive monitoring method for subway tunnel wall stress in this embodiment include:

[0056] Step S01. Monitor the vibration signals collected by the train vibration sensor and the optical signals collected by the distributed optical fiber sensor installed on the subway tunnel wall in real time. Demodulate the monitored optical signals to obtain stress signals, and extract the temperature change and Bragg wavelength change. Calculate the actual stress under thermo-mechanical coupling based on the extracted temperature change and Bragg wavelength change. Extract the dynamic displacement of the tunnel wall caused by train vibration based on the monitored vibration signals, and calculate the dynamic stress based on the dynamic displacement of the tunnel wall. Calculate the actual stress data under thermo-mechanical-vibration multi-field coupling based on the actual stress under thermo-mechanical coupling and the dynamic stress.

[0057] Step S101 Fiber Optic Sensor Data Acquisition and Demodulation

[0058] In this embodiment, distributed fiber optic sensors are installed on the subway tunnel wall to monitor stress signals in real time. The optical signals inside the fiber optics are collected in real time, and data collected over a long monitoring period is extracted from the fiber optic sensors. A fiber optic demodulator is then used to demodulate the information in the optical signals into stress signals. By using fiber optic sensing technology to survey the subway tunnel wall, full coverage of the tunnel wall can be easily achieved, providing long-term, continuous data support. This allows for convenient and sustained real-time stress monitoring throughout the entire tunnel structure, improving data comprehensiveness and real-time performance. This enables high-precision, real-time stress monitoring of the subway tunnel and also allows it to adapt to the complex environmental changes within the subway tunnel.

[0059] This embodiment constructs a stress monitoring system based on fiber optic sensing technology, enabling high-precision, real-time monitoring of subway tunnel walls. The system utilizes a distributed arrangement of fiber optic sensors within the tunnel structure to comprehensively and continuously capture stress change signals, making it particularly suitable for complex and dynamic tunnel environments. Because fiber optic sensing technology is insensitive to external electromagnetic interference and has the advantage of long-distance transmission, it can continuously collect high-quality data over long periods. Stress information is extracted using a fiber optic demodulator, ensuring the high accuracy and stability of the stress data, thus providing reliable foundational data for subsequent analysis.

[0060] This embodiment also includes a vibration sensor in the monitoring system for real-time monitoring of key parameters of train vibration, such as vibration frequency. Amplitude This is done by combining vibration data from vibration sensors to correct the stress values ​​calculated directly based on the fiber Bragg grating (FBG) principle.

[0061] Step S102. Stress Calculation

[0062] After demodulating the stress signal in step S101, further stress calculation is performed to convert it into stress data. In the actual operation of subway tunnels, the stress borne by the tunnel wall is not a single mechanical stress, but is affected by the coupling effect of multiple physical fields such as train vibration and temperature changes. In this embodiment, the multi-field coupling effect of thermo-mechanical-vibration is fully considered, and the actual stress under the multi-field coupling effect of thermo-mechanical-vibration is calculated based on the vibration signal and the real-time temperature change, so as to obtain the real-time monitored stress data.

[0063] In this embodiment, the calculation of the actual stress under the multi-field coupling of thermo-mechanical-vibration includes:

[0064] Step S121. Extract the dynamic displacement of the tunnel wall based on the collected vibration signal, and extract the real-time temperature change based on the demodulated stress signal;

[0065] Step S122. Calculate the actual stress under thermo-mechanical coupling based on the Bragg wavelength and the real-time temperature change;

[0066] Step S123. Calculate the dynamic strain based on the real-time dynamic displacement of the tunnel wall, and then calculate the dynamic stress caused by the train vibration based on the calculated dynamic strain.

[0067] Step S124. The actual stress under thermo-mechanical coupling and the dynamic stress caused by train vibration are superimposed to obtain the actual stress under thermo-mechanical-vibration multi-field coupling.

[0068] This embodiment considers the effects of multi-physics field coupling, such as train vibration and temperature changes, and incorporates the influence of temperature changes on the Bragg wavelength into stress calculation. This allows for a more accurate reflection of the actual stress of the tunnel wall under thermo-mechanical coupling. Simultaneously, the dynamic stress caused by train vibration is calculated based on the dynamic displacement of the tunnel wall caused by train vibration. Combining the actual stress under thermo-mechanical coupling and the dynamic stress caused by train vibration yields the final actual stress under the thermo-mechanical-vibration multi-field coupling. This approach comprehensively considers the thermo-mechanical-vibration multi-field coupling effect, effectively improving the accuracy of subway tunnel wall stress calculation and achieving precise calculation of subway tunnel wall stress.

[0069] To calculate the actual stress under the coupled effects of thermo-mechanical-vibration fields, this embodiment presents the following analysis:

[0070] First, we analyze the traditional process of calculating stress based on the principle of fiber Bragg gratings (FBG):

[0071] Bragg Wavelength With the period of fiber grating and effective refractive index The following relationship exists:

[0072] (1)

[0073] in, This indicates the grating period.

[0074] When a fiber Bragg grating is subjected to stress, its period and effective refractive index All of these will change, thus affecting the Bragg wavelength. The change in strain. According to the theory of elastic optics, the relationship between strain and changes in effective refractive index and Bragg wavelength is expressed as:

[0075] (2)

[0076] in, It is the change in the Bragg wavelength. It is the effective elastic-optical coefficient of the optical fiber, for example, for quartz optical fiber. The value is usually around 0.22. Indicates strain.

[0077] According to Hooke's Law in mechanics of materials, stress... With strain The following relationship exists:

[0078] (3)

[0079] in, It is the elastic modulus of the optical fiber, for example, for silica optical fiber. The value is approximately Pa.

[0080] The stress can then be obtained from the above formula. With Bragg wavelength change The relationship between them:

[0081] (4)

[0082] Then, the stress value calculated based on the fiber Bragg grating (FBG) principle is corrected, and the static stress σ′ calculated by combining the thermo-mechanical coupling model with the dynamic stress caused by train vibration is adjusted. The actual stress under the combined action of thermo-mechanical-vibration multi-field coupling is obtained by superposition.

[0083] During the actual operation of subway tunnels, the tunnel walls are affected by the coupling effects of multiple physical fields, such as train vibration and temperature changes. When the temperature changes, the effective refractive index and grating period of the optical fiber will change, which will lead to a Bragg wavelength shift. The temperature change ΔT will affect the effective refractive index and grating period. The change in effective refractive index caused by temperature change can be expressed as:

[0084] (5)

[0085] in, Thermo-optic coefficient, Indicates the effective refractive index of the optical fiber. Indicates due to temperature change The resulting change in effective refractive index.

[0086] The change in grating period due to temperature variation can be expressed as:

[0087] (6)

[0088] in, The coefficient of thermal expansion of the optical fiber material. Indicates the grating period, Indicates due to temperature change The resulting change in grating period It represents the amount of temperature change.

[0089] Combining the two formulas above, we can deduce the effect of temperature change. Caused Bragg wavelength change :

[0090] (7)

[0091] Therefore, we get:

[0092] (8)

[0093] According to the FBG principle, the change in Bragg wavelength caused by stress is: Then, by superimposing the Bragg wavelength changes caused by temperature changes and stress, the total Bragg wavelength change is obtained. + .

[0094] The original stress before thermo-mechanical coupling correction With Bragg wavelength change The relationship is as follows:

[0095] (9)

[0096] in, The effective optical-elastic coefficient is E, where E is the elastic modulus of the optical fiber. Indicates the effective refractive index.

[0097] The actual stress (static stress σ′) under thermo-mechanical coupling after correction. Total change in Bragg wavelength The relationship between them is:

[0098] (10)

[0099] in, This represents the actual stress under thermo-mechanical coupling. Indicates the effective elastic coefficient. Indicates the effective refractive index, This represents the elastic modulus of the optical fiber. This represents the total change in the Bragg wavelength. Indicates the Bragg wavelength. This represents the coefficient of thermal expansion of the optical fiber material. Indicates the thermo-optic coefficient. It represents the amount of temperature change.

[0100] This embodiment constructs the thermo-mechanical coupling model shown in the above formula (10). This model incorporates the influence of temperature change on Bragg wavelength into stress calculation, which can more accurately reflect the actual stress of the tunnel wall under thermo-mechanical coupling.

[0101] Further analysis of the dynamic stress caused by train vibration:

[0102] This embodiment establishes a vibration model that fits the structural characteristics of a subway tunnel based on structural dynamics theory, using a simplified single-degree-of-freedom vibration model as an example. Specifically, the dynamic displacement x(t) of the tunnel wall caused by train vibration can be expressed as:

[0103] (11)

[0104] Due to dynamic stress With dynamic strain Relatedly, dynamic strain is also connected to dynamic displacement. Assuming the elastic modulus of the tunnel wall material is E, the dynamic strain can be obtained through calculations such as differentiating the dynamic displacement. Then, the dynamic stress caused by train vibration can be calculated. :

[0105] (12)

[0106] in, This indicates the dynamic stress caused by train vibration. This represents the elastic modulus of the optical fiber. It represents dynamic strain.

[0107] Then the static stress calculated by the thermo-mechanical coupling model Dynamic stress caused by train vibration By superposition, the actual stress of the tunnel wall under the coupled action of multiple fields of thermo-mechanical-vibration is obtained. :

[0108] (13)

[0109] Therefore, the actual stress under the coupled action of thermo-mechanical-vibration fields can be obtained. The calculation expression is:

[0110] = + (14)

[0111] Furthermore, in the scenario of elliptical deformation of a subway tunnel coupled with train vibration, assuming the tunnel radius is R, the train vibration amplitude is A, the frequency is f, and the material Poisson's ratio is... , Let time be the time factor. When the train vibration causes the tunnel wall to deform into an elliptic shape, its radial displacement can be expressed as: ,in The angle is circumferential, which results in radial strain of... tangential strain is According to mechanics of materials, tangential stress Considering the circumferential symmetry of the elliptic deformation, the mean tangential stress is taken as:

[0112] (15)

[0113] In the scenario where train vibration causes elliptic deformation of the tunnel wall, the dynamic stress induced by the train vibration... That is, the average value of the tangential stress. :

[0114] (16)

[0115] Equation (15) above introduces R, A, f, It can quantify the coupling effect between tunnel elliptic deformation and train vibration, and can better match the dynamic stress characteristics of subway tunnels.

[0116] The average tangential stress After superimposing the thermo-mechanical coupled stress, the actual stress under the thermo-mechanical-vibration multi-field coupling effect of the elliptic deformation of the tunnel in this scenario can be obtained. :

[0117] (17)

[0118] like Figure 2As shown, this embodiment acquires the optical signal collected by an optical fiber sensor installed on the subway tunnel wall, demodulates the optical signal using an optical fiber demodulator, and calculates the actual stress value according to the Bragg wavelength change, temperature change, and vibration data using the above formula (14) or (17). By adopting the above method, the actual stress under the multi-field coupling of thermo-mechanical-vibration is constructed. The computational model can comprehensively consider the multi-field coupling effect of thermo-mechanical-vibration, and achieve accurate calculation of the stress in the subway tunnel wall.

[0119] Step S02. The stress signal monitored in real time by the distributed optical fiber sensor is denoised using wavelet transform. The denoised stress signal is used as the observation data, and the actual stress data under the multi-field coupling of thermo-mechanical-vibration is used as the initial stress state data. Kalman filtering is used for prediction to obtain the corrected thermo-mechanical-vibration coupled stress data.

[0120] In tunnel stress monitoring, the acquired signals are often affected by environmental noise, equipment interference, and other factors, resulting in significant noise levels. This is especially true in complex environments where temperature changes and mechanical interference can lead to signal errors and distortion. While traditional signal processing methods can denoise, their effectiveness is limited and they are prone to introducing processing errors. The Kalman filter, through a recursive algorithm, progressively optimizes the signal estimate, effectively removing high-frequency noise. Wavelet transform, through multi-resolution analysis, can separate the noise component from the true signal, further improving signal quality. This embodiment employs a combination of wavelet transform and Kalman filtering to denoise the wavelength-varying signals obtained from fiber optic sensors. First, wavelet transform removes high-frequency noise. Kalman filtering then further smooths the signal, preventing interference from excessively strong high-frequency noise. Simultaneously, wavelet transform separates the high-frequency and low-frequency components of the signal, and Kalman filtering further optimizes the signal using a model. This not only extracts the effective signal and effectively removes high-frequency noise but also optimizes the signal through Kalman filtering, improving signal quality and accuracy.

[0121] In this embodiment, the stress data monitored in real time by the distributed fiber optic sensor is first subjected to noise reduction processing using wavelet transform. The steps include:

[0122] Step S201. Select a wavelet basis based on the frequency characteristics and noise distribution of the stress data. Use the selected wavelet basis to perform multi-layer wavelet decomposition on the stress data monitored in real time by the distributed optical fiber sensor. Use a preset dynamic threshold to remove high-frequency noise from the low-frequency and high-frequency coefficients to obtain the denoised low-frequency and high-frequency coefficients. After inverse wavelet transformation, obtain the first denoised stress data.

[0123] Step S202. The stress data after the first denoising is processed using an adaptive filter to remove low-frequency noise, and the weights of the filter are continuously adjusted to adaptively track the changes in low-frequency noise, so as to obtain the stress data after the second denoising.

[0124] Thermal coupling can introduce low-frequency noise into the signal, which may be related to temperature fluctuations and train vibrations. Meanwhile, fiber optic demodulation and environmental interference can also introduce high-frequency noise. Low-frequency noise may mask the true stress variation trend of the tunnel wall, while high-frequency noise can interfere with the detailed characteristics of the signal. Vibration sensors, when acquiring train vibration signals, are also affected by environmental vibrations and their own noise. In this embodiment, during the noise reduction process, a wavelet basis with better time-frequency localization characteristics is first selected based on the signal's frequency characteristics and noise distribution. Wavelet transform is then used to decompose the signal into multiple scales, extracting low-frequency and high-frequency components. A threshold denoising method (such as soft thresholding) is then used to remove high-frequency noise, effectively preserving the main signal features while reducing the impact of noise on stress prediction. Simultaneously, considering the low-frequency noise characteristics caused by temperature changes and train vibrations, an adaptive filter is used to remove low-frequency noise, continuously adjusting the filter weights to adaptively track changes in low-frequency noise, effectively removing it. Then, Kalman filtering further optimizes the signal, eliminating residual noise through prediction and update steps, improving signal accuracy and stability. These noise reduction and optimization processes together ensure higher quality stress data and further optimize the accuracy and reliability of the stress signal. This ensures stable and accurate stress data can be obtained even in complex environments, meeting the requirements of real-time monitoring.

[0125] In specific application embodiments, such as Figure 3 As shown, the specific implementation steps for wavelet transform of the stress signal monitored in real time by the distributed optical fiber sensor in this embodiment include:

[0126] Step S211. Select a suitable wavelet basis and decompose it.

[0127] Based on the frequency characteristics and noise distribution of the signal, this embodiment selects a wavelet basis with better time-frequency localization characteristics. The selected wavelet basis is used to decompose the signal into a low-frequency (approximation) part and a high-frequency (detail) part. Multi-scale features of the signal are extracted through multi-level wavelet decomposition.

[0128] Assuming the stress signal is monitored in real time by a distributed fiber optic sensor. For a one-dimensional signal, wavelet basis functions are used at each level of decomposition. and scaling function Convolve the signal, where:

[0129] Low-frequency component coefficients (approximation): obtained by applying the input stress signal. With scaling function (Translation and scaling of the mother wavelet) are used for convolution to obtain the low-frequency coefficients. : ;

[0130] High-frequency coefficients (details): obtained by applying the input stress signal and wavelet basis functions. Convolution is performed using the translation and scaling of the mother wavelet to obtain the high-frequency coefficients. : ;

[0131] The stress signal was processed in the manner described above. Multi-level wavelet decomposition is performed to obtain approximations and details at multiple different scales. The scale of the multi-level wavelet decomposition is determined by the number of decomposition levels (e.g., a first-level decomposition corresponds to scale 2¹, a second-level decomposition to scale 2²), by using the stress signal... The decomposition is divided into a low-frequency approximation part Aj and a high-frequency detail part Dj in different frequency bands. The specific scale can be determined by selecting a wavelet basis based on the frequency characteristics of the stress signal and the noise distribution. Each level of decomposition further decomposes the low-frequency part, that is, the next level further decomposes the low-frequency coefficients obtained from the previous level. For example, the first-level decomposition yields... and Then to Perform a second-level decomposition to obtain And so on.

[0132] Step S212. Threshold denoising

[0133] This embodiment effectively removes high-frequency noise by thresholding the high-frequency coefficients. Specifically for tunnel fiber optic monitoring scenarios, when using a preset dynamic threshold to denoise the high-frequency coefficients in each layer decomposition, if the absolute value of the coefficient corresponding to the effective signal exceeding a specified proportion is greater than the dynamic threshold T, or the absolute value of the coefficient corresponding to the noise exceeding a specified proportion is mostly less than the dynamic threshold T (i.e., the absolute value of the effective signal coefficient is mostly greater than T or the absolute value of the noise coefficient is mostly less than the dynamic threshold T), it indicates that the effective signal and noise can be clearly distinguished in amplitude by the dynamic threshold T. In other words, the boundary between high-frequency noise and effective signal is clear, and a hard thresholding method is used to directly filter out noise below the threshold. Otherwise, if the noise and effective signal spectra overlap significantly, a soft thresholding method is used to smooth the high-frequency coefficients to preserve signal details.

[0134] When using the hard threshold method:

[0135] ;

[0136] ;

[0137] in, It is the first j High-frequency coefficients (detail coefficients) after layer wavelet decomposition. This represents the high-frequency coefficients after noise reduction processing. This is a preset dynamic threshold. If the absolute value of a coefficient is greater than the set threshold, the coefficient is retained; if it is less than the threshold, it is set to zero. Dynamic Threshold It can be obtained by setting it up as follows:

[0138] (18)

[0139] in, For the first Standard deviation of layer high-frequency coefficients (detail coefficients), For the first The length of the high-frequency coefficient of the layer, This is the tunnel vibration and noise impact factor (for example, it can be taken as 0.1-0.3).

[0140] As shown in equation (18), this embodiment constructs a dynamic threshold based on the extreme value statistical theory of Gaussian noise. Let's assume the first The noise component of the layer wavelet detail coefficients follows a zero-mean Gaussian distribution, combined with the coefficient length. The theoretical threshold term for a purely noisy scene is obtained from the extreme value asymptotic properties. However, the detail factor of tunnel stress contains both effective signals (thermal-mechanical-vibration coupling characteristics) and noise. The signal can raise extreme values, and vibration noise needs to be dynamically adapted. Therefore, a signal-noise coupling correction term is introduced. Finally, the theoretical threshold is multiplied by the dynamic correction term to construct a dynamic threshold that integrates noise statistics, signal energy characteristics, and environmental interference features. It can achieve dynamic threshold adjustment by fusing noise intensity and signal mean, thereby further improving the accuracy and robustness of denoising.

[0141] When using the soft threshold method:

[0142] (19)

[0143] The soft thresholding method is similar to the hard thresholding method, when the first... The high frequency coefficient of the layer is greater than the dynamic threshold. T ,Right now ,according to We obtain the denoised high-frequency coefficients, but when the absolute value of the high-frequency coefficients... When it is less than the dynamic threshold T, that is The high-frequency coefficients of the current layer are reduced in size and then used as the high-frequency coefficients after denoising. Instead of setting it directly to zero.

[0144] After the thresholding process described above, the denoised low-frequency and high-frequency coefficients are reconstructed back to the original signal using the inverse wavelet transform (IWT). This assumes that the signal has already undergone wavelet decomposition to obtain low-frequency and high-frequency coefficients at different levels. The transformation combines these coefficients to recover an approximate version of the original signal: ).

[0145] The stress signal underwent preliminary noise reduction using the wavelet transform described above, primarily removing high-frequency noise. However, some low-frequency noise or incompletely removed abnormal fluctuations may still exist. Therefore, Kalman filtering was further used to optimize the signal. Figure 4 As shown, through the prediction and update steps of Kalman filtering, the accurate signal is extracted from the noise interference. In this embodiment, the actual stress data under the force-vibration multi-field coupling calculated in step S01 is first used as the initial state estimate, and Kalman filtering is used for prediction. The expression is:

[0146] (20)

[0147] (twenty one)

[0148] in, This is the predicted stress state at the current moment, and at the initial moment... This refers to the actual stress data obtained in step S01 under the combined action of thermo-mechanical-vibration multi-fields. It is the covariance matrix of the state estimate. It is the process noise covariance matrix.

[0149] After prediction, the denoised stress signal is used as the observation data, and the actual observation value is combined with the predicted value to update the current state estimate. The expression for state update using Kalman filtering is as follows:

[0150] (twenty two)

[0151] (twenty three)

[0152] (twenty four)

[0153] in, Kalman gain is used to weigh the "confidence" between predicted and observed values. It measures the noise covariance matrix. It is the updated state estimate, i.e., the corrected thermo-mechanical-vibrational coupled stress data.

[0154] Assuming a simple linear model is used, the state transition matrix A=1, meaning the current state is equal to the state at the previous time step (assuming the state changes linearly), and the observation matrix H=1, meaning each observation directly reflects the current state. For the first time step (k=1), there is no estimate of the state at the previous time step, so the initial estimate can be used directly.

[0155] ;

[0156] Prediction error covariance:

[0157] ;

[0158] for The time, assuming it is known. ,as well as ,therefore:

[0159] ;

[0160] Update the current state estimate based on the predicted state and the Kalman gain:

[0161] ;

[0162] That is, the updated state estimate at time 1 is 1.71875, and the update error covariance is:

[0163] .

[0164] Furthermore, considering the existence of nonlinear dynamic equations in the tunnel stress system, such as the signal vibration amplitude exceeding the material's elastic limit, or the existence of ill-conditioned observation matrices (e.g., the Jacobian matrix condition number), ), or there is non-stationary noise (e.g., noise covariance ratio). For example, the Extended Kalman Filter (EKF) is used for prediction. Based on the core covariance update formula of the Kalman Filter, the Extended Kalman Filter is locally linearized to make it applicable to nonlinear, ill-conditioned, and non-stationary tunnel stress systems.

[0165] Specifically, in this embodiment, the calculation expression for prediction using the Extended Kalman Filter (EKF) is as follows:

[0166] State prediction: ;

[0167] Error covariance prediction: ;

[0168] Kalman gain: ;

[0169] Status Update: ;

[0170] Error covariance update: ;

[0171] in, The Kalman gain at time k is... Let $\mathbf{k}$ be the covariance of the stress state after the update at time $k$. Let $\mathbf{k}$ be the covariance of the predicted stress state at time $k$. For the observation matrix, The value at time k is the observed value, which is the stress signal obtained after wavelet transform and noise reduction from the stress signal monitored in real time by the distributed fiber optic sensor at time k. To observe the noise covariance, To predict the stress state at time k, the initial time is... This refers to the actual stress data obtained in step S01 under the combined action of thermo-mechanical-vibration multi-fields. It is the identity matrix. This is the updated stress state estimate at time k, i.e., the corrected thermo-mechanical-vibrational coupled stress data. and It is the Jacobian matrix, used for linearizing nonlinear functions.

[0172] Based on the above method, the observation data at time 2 and time 3 are calculated sequentially. Through the prediction and update process of Kalman filtering, the stress value at each time is estimated step by step. As the amount of observation data increases, the confidence in the state estimation gradually increases (the error covariance gradually decreases), making the estimated value more reliable than the simple observation value, especially under the condition of large noise.

[0173] This embodiment combines multi-level wavelet transform with Kalman filtering for noise reduction, which can give full play to the advantages of each and effectively reduce the noise component in the signal. Wavelet transform can decompose the signal into components of different scales, and threshold denoising removes high-frequency noise while retaining the main information of the signal. Kalman filtering further optimizes the signal estimation through prediction and updating steps, reduces measurement errors, and improves the accuracy and stability of the signal. It is especially suitable for high-noise environments and can ensure the accuracy of stress monitoring data.

[0174] The stress data processing in this embodiment can be divided into the following three stages:

[0175] ① Theoretical Prior: First, in step S01, the actual stress data under the thermo-mechanical-vibration multi-field coupling effect is calculated using the multi-field coupling model as shown in equation (14) or (17) and used as theoretical prior data. This actual stress data under the thermo-mechanical-vibration multi-field coupling effect is used as the initial state data predicted by Kalman filter, i.e., the multi-field coupling model. To achieve physical constraints;

[0176] ② Observation purification: The optical signals collected in real time by the distributed optical fiber sensor are processed by wavelet transform to reduce noise, and the resulting data is used as the observation data, i.e., the raw optical fiber data. This enables denoised observation and noise suppression.

[0177] ③ Fusion Correction: The actual stress data under the coupled action of thermo-mechanical-vibration fields is used as the initial state data for Kalman filter prediction, and the denoised stress data is then used for Kalman filter prediction. To achieve optimal stress estimation, corrected thermo-mechanical-vibration coupled stress data are obtained. , to serve as input data for subsequent stress prediction models;

[0178] This embodiment can fuse the "predicted stress state" with the "actual observed value" through the above method, effectively correcting errors and obtaining more reliable stress estimation data. .

[0179] Due to the introduction of train vibration sensors, this embodiment fuses the vibration signals collected by the vibration sensors with the signals from the fiber optic sensors before noise reduction processing. By time synchronization, the data from different types of sensors are aligned in the time dimension, and then a multi-dimensional data vector containing wavelength change signals and vibration signals is constructed as the input for subsequent noise reduction processing.

[0180] Specifically, after acquiring the vibration signal collected by the train vibration sensor and the optical signal collected in real time by the distributed optical fiber sensor, and before noise reduction processing, this embodiment also includes a preprocessing step of fusing the vibration signal from the train vibration sensor and the optical signal from the distributed optical fiber sensor, including:

[0181] The vibration signal from the train vibration sensor is synchronized with the optical signal from the distributed optical fiber sensor in time to construct a multi-dimensional data vector containing wavelength change signals and vibration signals.

[0182] After initial denoising of the multidimensional data vector, it is input into a pre-trained denoising model to obtain a multi-field coupled signal after collaborative denoising. The denoising model is trained using a training set including vibration signals from train sensors and optical signals from fiber optic sensors, based on a deep learning model to learn noise features and real signal features in the signal.

[0183] Specifically, this embodiment employs different processing methods for different types of noise. Given the low-frequency noise characteristics caused by temperature changes and train vibrations, an adaptive filtering technique (e.g., an LMS filter) is used. By continuously adjusting the filter weights, it adaptively tracks changes in low-frequency noise, effectively removing it. For high-frequency noise processing, the wavelet basis selection is optimized during the wavelet transform stage. Based on the signal's frequency characteristics and noise distribution, a wavelet basis with better time-frequency localization characteristics is selected. Before denoising, the fiber optic sensor signal and vibration sensor signal are fused and preprocessed. Time synchronization technology ensures consistency between the two types of sensor data in the time dimension, thereby constructing a multi-dimensional data vector. Dimensionality reduction is then applied to this multi-dimensional data to extract key features, reduce data redundancy, and improve denoising efficiency. Finally, a joint denoising algorithm is used, leveraging the complementary information from the two types of sensor data. The fiber optic sensor signal and vibration sensor signal, after initial denoising, are input into a deep learning-based denoising model. Taking a CNN model as an example, by constructing multiple convolutional and pooling layers, the noise and real signal features in the signal are automatically learned, achieving collaborative denoising of multi-field coupled signals.

[0184] Step S03. Input the corrected thermo-mechanical-vibration coupled stress data into the pre-trained stress prediction model to obtain the final stress prediction result of the subway tunnel wall. The stress prediction model is obtained by training the LSTM-SVM model using a training dataset composed of historical stress data. The LSTM-SVM model uses the LSTM model to obtain the preliminary stress prediction result, and uses the output of the LSTM model as the input of the SVR model. The SVR model corrects the output result of the LSTM model to obtain the final stress prediction result.

[0185] After signal denoising, a relatively clean stress signal is obtained. This embodiment establishes a stress prediction model based on the stress signal, using historical data and current real-time data to accurately predict stress. The task of predicting stress signals typically faces challenges related to time-series dependence and nonlinear characteristics. LSTM, as a deep learning model, can effectively capture temporal patterns in data, while SVR excels at handling nonlinear relationships. Due to the nonlinear, time-series dependent, and noisy characteristics of tunnel stress signals, this embodiment combines Long Short-Term Memory (LSTM) and Support Vector Regression (SVR) (LSTM-SVR) to complete the prediction, fully leveraging the advantages of both while overcoming their respective limitations. LSTM, by learning the time-series characteristics of stress changes, can accurately predict future stress trends, while SVR performs nonlinear optimization based on the LSTM output, further improving prediction accuracy. By fusing LSTM and SVR, the system can automatically adjust and optimize prediction results according to actual conditions, not only improving the accuracy of stress prediction but also enhancing the robustness and adaptability of the system. This improves the model's accuracy, robustness, and applicability, enabling it to cope with complex stress changes in the tunnel environment.

[0186] In this embodiment, the specific steps for inputting the corrected thermo-mechanical-vibration coupled stress data into the pre-trained stress prediction model to obtain the stress prediction results for the subway tunnel wall include:

[0187] Step S301. Input the real-time denoised stress data into the LSTM model that has been trained using historical stress data. The LSTM model then predicts the stress changes in the future based on the historical stress data to obtain the stress prediction result.

[0188] Step S302. Use the output of the LSTM model as the input feature of the SVR model, and perform nonlinear optimization on the stress prediction value output by the SVR model to obtain the optimized prediction result.

[0189] Step S303. Fuse the output of the LSTM model with the output of the SVR model to obtain the fused prediction result;

[0190] Step S304. Obtain the final stress prediction result based on the optimized prediction result and the fusion prediction result.

[0191] In this embodiment, the fusion strategy for combining the output of the LSTM model and the output of the SVR model can be as follows:

[0192] Direct fusion method: The output of the LSTM model is used as the input feature of the SVR model, and the SVR model outputs the fused prediction result;

[0193] Multi-stage fusion method: The output of the LSTM model is used as the initial prediction result, and the output of the LSTM model is corrected by the SVR model to obtain the fused prediction result;

[0194] Weighted fusion method: The outputs of the LSTM model and the SVR model are weighted with different weights to obtain the fused prediction result.

[0195] The above-mentioned fusion strategy can be one of them, or two or more can be combined according to actual needs. It can also adaptively adjust the model weights according to different stress change modes to better cope with complex tunnel stress changes.

[0196] In the process of establishing the stress prediction model, this embodiment combines LSTM and SVR models to achieve stress prediction. This fully leverages the advantages of both models in time-dependent and nonlinear regression optimization. LSTM captures the time dependence of tunnel stress data, accurately modeling the stress change trend. SVR performs nonlinear optimization based on LSTM, further improving the prediction accuracy and robustness. The fusion strategy can also effectively combine the advantages of LSTM and SVR, reducing their respective limitations, thereby establishing an accurate and robust stress prediction model, achieving more precise stress prediction, and meeting the actual needs of subway tunnel stress monitoring.

[0197] In specific application embodiments, such as Figure 5 As shown, the detailed steps for training and predicting the LSTM-SVR model are as follows:

[0198] (1) LSTM model training

[0199] The basic architecture of LSTM includes: input layer → LSTM layer → fully connected layer → output layer;

[0200] Specifically, the input layer receives the thermo-mechanical-vibrational coupled stress data after Kalman filtering correction, which is symbolically represented as: ; yes The optimal stress estimate of the Kalman filter output at time t is the corrected thermo-mechanical-vibration coupled stress data.

[0201] The LSTM layer constructs the core layer by stacking multiple LSTM units to capture the long-term dependencies of stress time series. The multi-layer extension enhances the model's ability to learn non-stationary multi-field coupling features.

[0202] Since tunnel stress prediction is a continuous value regression problem (non-classification), the output layer preferentially uses a linear activation function (to avoid the interval compression error of the Sigmoid function), the formula is as follows:

[0203] (25)

[0204] in, for Predicted values ​​of thermo-mechanical-vibration coupled stress at any given time. The hidden state output by the LSTM layer. These represent the weights and biases of the fully connected layer.

[0205] LSTM training process: Select a loss function suitable for the regression problem, including mean squared error (MSE) or root mean square error (RMSE), use the Adam optimizer to update the model weights through backpropagation, divide the data into training and validation sets, evaluate the performance of the LSTM model through cross-validation or validation set, and adjust the model hyperparameters.

[0206] The LSTM model will predict stress changes at future moments based on the input historical data.

[0207] (2) SVR optimization and fusion

[0208] The output of the LSTM model is used as the input feature of the SVR. After learning temporal dependencies, the LSTM provides a preliminary prediction result based on historical information, which the SVR then further optimizes. The SVR effectively captures nonlinear features through kernel methods, and by selecting appropriate parameters, it ensures good tolerance to noise and achieves good predictive performance in complex stress modes.

[0209] The data is divided into training and testing sets. The SVR model is trained using the training set and its performance is evaluated using the testing set. Based on the output of the LSTM and its nonlinear optimization, the SVR generates more accurate stress prediction values.

[0210] (3) Fusion of LSTM and SVR

[0211] This embodiment forms an LSTM → SVR structure. The LSTM model first generates time series predictions based on the input historical data, and the SVR receives the output of the LSTM. The LSTM-SVR model then... To train the input, learn the temporal patterns, and predict future stress values. The prediction is further optimized using nonlinear regression to obtain the output result. For example, the following decision rule can be used to achieve early warning of multi-field coupled stress over-limit through dynamic thresholds:

[0212] (26)

[0213] in, This represents the judgment indicator. 1 and 0 represent binary decision results. For example, 1 indicates that an over-limit warning is triggered, and 0 indicates that an over-limit warning is not triggered. These are the stress prediction values ​​output by the LSTM-SVR model. The baseline threshold can be any one of the standardized value, simulated value, or statistical value, or two or more fused values ​​can be used. The tunnel environment sensitivity coefficient (can be taken as 0.5~1.0). The standard deviation of historical stress.

[0214] In this embodiment, the LSTM is responsible for capturing time dependencies and the SVR is responsible for refining nonlinear patterns, thus fully integrating the LSTM and SVR models to obtain more accurate prediction results.

[0215] This project utilizes fiber optic sensing technology to monitor tunnel stress throughout its entire lifecycle, employing multi-level processing including signal demodulation, noise reduction, and optimization to ensure high-precision stress data. Secondly, by combining LSTM and SVR prediction models, it fully considers the temporal characteristics of tunnel stress signals while effectively handling nonlinear variations. This multi-level signal processing and prediction mechanism enables full coverage of the tunnel structure, significantly improving data comprehensiveness and real-time performance. Furthermore, traditional signal processing methods may fail to effectively separate noise from signals in noisy environments, leading to significant measurement errors. By combining wavelet transform and Kalman filtering, this scheme effectively denoises and optimizes the signal, improving data accuracy. Finally, existing stress prediction techniques are often limited by model capabilities when dealing with complex, nonlinear stress changes, resulting in inaccurate predictions. By combining LSTM and SVR, this scheme fully leverages the advantages of both, improving prediction accuracy and robustness, particularly in the context of rapid and complex stress changes in tunnel environments, significantly enhancing stress prediction accuracy.

[0216] This embodiment further provides a subway tunnel wall stress adaptive monitoring device, including a processor and a memory. The memory is used to store a computer program, and the processor is used to execute the computer program to perform the method described above.

[0217] It is understood that the method described in this embodiment can be executed by a single device, such as a computer or server, or it can be applied to a distributed scenario where multiple devices cooperate to complete the task. In a distributed scenario, one of the multiple devices may execute only one or more steps of the method described in this embodiment, and the multiple devices interact to complete the method. The processor can be implemented using a general-purpose CPU, microprocessor, application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the method described in this embodiment. The memory can be implemented using read-only memory (ROM), random access memory (RAM), static storage devices, and dynamic storage devices. The memory can store the operating system and other applications. When the method described in this embodiment is implemented through software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0218] This embodiment further provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0219] Those skilled in the art will understand that the above embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce implementations of the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0220] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims

1. An adaptive monitoring method for stress in subway tunnel walls, characterized by the following steps: include: The system monitors vibration signals collected by train vibration sensors and optical signals collected in real time by distributed fiber optic sensors installed on the subway tunnel wall. After demodulating the monitored optical signals, stress signals are obtained, and temperature and Bragg wavelength changes are extracted. The actual stress under thermo-mechanical coupling is calculated based on these extracted values. The dynamic displacement of the tunnel wall caused by train vibration is extracted from the monitored vibration signals, and dynamic stress is calculated based on this dynamic displacement. Finally, the actual stress data under thermo-mechanical-vibration multi-field coupling is calculated based on the actual stress under thermo-mechanical coupling and the dynamic stress. When train vibration induces elliptic deformation of the tunnel wall, the dynamic stress caused by the train vibration is... Mean tangential stress : ， in, Where is the tunnel radius. For train vibration amplitude, For frequency, For the material's Poisson's ratio, For time, This represents the elastic modulus of an optical fiber. The stress signal monitored in real time by the distributed optical fiber sensor is denoised using wavelet transform. The denoised stress signal is used as the observation data, and the actual stress data under the thermo-mechanical-vibration multi-field coupling is used as the initial stress state data. Kalman filtering is used for prediction to obtain the corrected thermo-mechanical-vibration coupled stress data. The corrected thermo-mechanical-vibration coupled stress data is input into a pre-trained stress prediction model to obtain the final stress prediction result of the subway tunnel wall. The stress prediction model is obtained by training an LSTM-SVM model using a training dataset composed of historical stress data. The LSTM-SVM model uses the LSTM model to obtain the initial stress prediction result, and uses the output of the LSTM model as the input of the SVR model. The SVR model corrects the output result of the LSTM model to obtain the final stress prediction result.

2. The adaptive monitoring method for subway tunnel wall stress according to claim 1, characterized in that, The actual stress data under thermo-mechanical-vibration multi-field coupling calculated based on the actual stress under the thermo-mechanical coupling and the dynamic stress includes: The dynamic displacement of the tunnel wall is extracted from the collected vibration signals, and the real-time temperature change and Bragg wavelength change are extracted from the demodulated stress signals. The actual stress under thermo-mechanical coupling is calculated based on the change in Bragg wavelength and the real-time temperature change. Dynamic strain is calculated based on the real-time dynamic displacement of the tunnel wall, and then the dynamic stress caused by train vibration is calculated based on the calculated dynamic strain. The actual stress under the thermo-mechanical coupling effect and the dynamic stress caused by the train vibration are superimposed to obtain the actual stress under the thermo-mechanical-vibration multi-field coupling effect.

3. The adaptive monitoring method for subway tunnel wall stress according to claim 2, characterized in that, The calculation expression for the actual stress under the aforementioned thermo-mechanical coupling effect is as follows: ， in, This represents the actual stress under thermo-mechanical coupling. Indicates the effective elastic coefficient. Indicates the effective refractive index, This represents the elastic modulus of the optical fiber. This represents the total change in the Bragg wavelength. Indicates the Bragg wavelength. This represents the coefficient of thermal expansion of the optical fiber material. Indicates the thermo-optic coefficient. It represents the amount of temperature change.

4. The adaptive monitoring method for subway tunnel wall stress according to claim 3, characterized in that, The formula for calculating the dynamic stress caused by train vibration is as follows: ， in, This indicates the dynamic stress caused by train vibration. This represents the elastic modulus of the optical fiber. Indicates dynamic strain; Actual stress under thermo-mechanical-vibration multi-field coupling The calculation expression is: = + 。 5. The adaptive monitoring method for subway tunnel wall stress according to claim 1, characterized in that, The noise reduction processing of the stress data monitored in real time by the distributed fiber optic sensor using wavelet transform includes: Based on the frequency characteristics and noise distribution of the stress data, a wavelet basis is selected. The selected wavelet basis is used to perform multi-level wavelet decomposition on the stress signal monitored in real time by the distributed optical fiber sensor. In each decomposition, the low-frequency coefficients are obtained by convolving the input stress signal with the scaling function and convolving the input stress signal with the wavelet basis function. The next level further decomposes the low-frequency coefficients obtained from the previous level. In each decomposition, the high-frequency coefficients are denoised using a preset dynamic threshold to remove high-frequency noise. The dynamic threshold is dynamically set according to the parameters of the high-frequency coefficients in each level. Finally, the low-frequency coefficients obtained after multi-level wavelet decomposition and the denoised high-frequency coefficients are subjected to inverse wavelet transform to obtain the denoised stress data. The stress data after the first denoising is processed by an adaptive filter to remove low-frequency noise, and the filter weights are continuously adjusted to adaptively track the changes in low-frequency noise, resulting in stress data after the second denoising.

6. The adaptive monitoring method for subway tunnel wall stress according to claim 5, characterized in that, The step of using a preset dynamic threshold to denoise the high-frequency coefficients in each layer of decomposition to remove high-frequency noise includes: If the absolute value of the coefficient corresponding to an effective signal exceeding a specified proportion in the current high-frequency coefficients is greater than the dynamic threshold, or the absolute value of the coefficient corresponding to noise exceeding a specified proportion is significantly less than the dynamic threshold, then a hard thresholding method is used based on the dynamic threshold; otherwise, a soft thresholding method is used based on the dynamic threshold. If a hard thresholding method is used... hour, , , Indicates the first j High-frequency coefficients after layer wavelet decomposition This represents the high-frequency coefficients after noise reduction processing; If a soft thresholding method is used for processing... hour, , At that time, the high-frequency coefficients of the current layer will be... The high-frequency coefficients after being reduced in size are used as the noise reduction coefficients. ; Dynamic threshold The calculation expression is: ， in, For the first Standard deviation of the high frequency coefficient of the layer For the first The length of the high-frequency coefficient of the layer, This refers to the factors affecting tunnel vibration and noise.

7. The adaptive monitoring method for subway tunnel wall stress according to any one of claims 1 to 6, characterized in that, Using the denoised stress signal as observation data and the actual stress data under the thermo-mechanical-vibration multi-field coupling as the initial stress state data, an extended Kalman filter is used for prediction, including: according to Perform state prediction, according to To predict the error covariance, according to The Kalman gain was calculated. ,according to Perform a status update, according to Perform error covariance update, where, The Kalman gain at time k is... Let $\mathbf{k}$ be the covariance of the stress state after the update at time $k$. Let $\mathbf{k}$ be the covariance of the predicted stress state at time $k$. For the observation matrix, The value at time k is the observed value, which is the stress signal obtained after wavelet transform and noise reduction from the stress signal monitored in real time by the distributed fiber optic sensor at time k. To observe the noise covariance, To predict the stress state at time k, the initial time is... This refers to the actual stress data under the combined effects of thermo-mechanical-vibration multi-field coupling. It is the identity matrix. The updated stress state estimate at time k is the corrected thermo-mechanical-vibrational coupled stress data. and These are Jacobian matrices, respectively.

8. The adaptive monitoring method for subway tunnel wall stress according to any one of claims 1 to 6, characterized in that, The step of inputting the corrected thermo-mechanical-vibration coupled stress data into a pre-trained stress prediction model to obtain the final stress prediction result of the subway tunnel wall includes: The corrected thermo-mechanical-vibration coupled stress data is input into an LSTM model that has been trained using historical stress data. The LSTM model then predicts the stress changes at future moments based on the historical stress data to obtain the stress prediction result. The output of the LSTM model is used as the input feature of the SVR model. The SVR model performs nonlinear optimization on the stress prediction value output by the LSTM model to obtain the optimized prediction result. The outputs of the LSTM model and the SVR model are fused to obtain the fused prediction result; The final stress prediction result is obtained based on the optimized prediction result and the fused prediction result.

9. A subway tunnel wall stress adaptive monitoring device, comprising a processor and a memory, wherein the memory is used to store a computer program, characterized in that, The processor is used to execute the computer program to perform the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images