Identification and positioning method for non-uniform deformation of subsea tunnel structure across fault zone

Abstract

The invention discloses a cross-fault-zone subsea tunnel structure non-uniform deformation identification and positioning method, which comprises the following steps: firstly, acquiring cross-fault-zone subsea tunnel BOTDA strain data, and respectively carrying out abnormal value processing, fitting and tidal residual strain separation on the strain data to obtain strain data with fault zone spatial characteristics; secondly, the strain data of the fault zone and the adjacent non-fault zone are clustered based on a Gaussian mixture model, a clustering strain mean value is extracted, and a strain mean value difference feature vector is constructed; a PSO algorithm is adopted to optimize an RF model to recognize a strain mean value difference time-varying curve, an LDSA algorithm is adopted to generate a high-resolution three-dimensional feature thermodynamic diagram for a non-uniform deformation region, and therefore the non-uniform deformation region is accurately positioned. According to the method, solid data support can be provided for identification, evaluation and disposal of uneven deformation of the tunnel structure, so that a more scientific management and maintenance decision is made, and subway operation safety is effectively guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of subway tunnel operation monitoring and relates to a method for identifying uneven deformation during the operation of a cross-fault zone submarine tunnel. Background Technology

[0002] Accurately identifying uneven deformation during subway tunnel operation is crucial for ensuring the structural safety of subways. In areas with poor geological conditions, tunnel structures are prone to uneven stress due to external loads, which can lead to a series of chain reactions such as segment joint separation and reduced track smoothness, resulting in varying degrees of uneven deformation in the tunnel and seriously threatening train operation safety. Summary of the Invention

[0003] To address the challenge of identifying non-uniform deformation in complex environments such as submarine tunnels spanning fault zones, this invention provides a method for identifying and locating non-uniform deformation in such tunnels. This method employs BOTDA distributed fiber optic sensing technology to monitor the strain of the subway tunnel cross-section. Based on the acquired strain data from numerous monitoring points, it diagnoses the safety status of the tunnel structure and precisely locates the positions of non-uniform deformations. This eliminates safety hazards caused by non-uniform deformation during the operation of subway tunnels spanning fault zones, thereby ensuring the operational safety of the subway and playing a crucial role in ensuring the long-term safe operation of subway tunnels.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A method for identifying and locating non-uniform deformation in a submarine tunnel structure spanning a fault zone includes the following steps:

[0006] Step 1: Preprocessing of strain data for submarine tunnels crossing fault zones:

[0007] Step 11: Deploy BOTDA distributed optical fiber in the subway undersea tunnel, collect BOTDA strain signals, and perform outlier processing and normalization on the collected vibration signals. Use regression method to fit the missing values ​​of the strain data.

[0008] Step 1 and Step 2: The Savzky-Golay (SG) algorithm is used to reduce noise in the strain data processed in Step 1, thus realizing the preprocessing of the strain data;

[0009] Step 2: Tidal residual strain separation:

[0010] Step 21: Extract ocean tidal data for the same time period in the target area and decompose the tidal data to extract the amplitude and phase parameters of the main astronomical tides:

[0011]

[0012] in, The number of tidal constituents refers to the total number of astronomical tidal constituents included in the analysis. Tidal height; Mean sea level; For tide division The average amplitude; For tide division angular frequency; For tide division The phase delay angle; For astronomical parameter nodal factors; It is the astronomical phase angle; For residual terms;

[0013] Step 22: Based on least squares spectral analysis, the main tidal components under tidal load are separated. The objective function is:

[0014]

[0015] Steps 2 and 3: Quantitatively analyze the correlation between tunnel strain and tidal level using Pearson correlation analysis.

[0016] Step 24: Using the cross-correlation function method, quantitatively characterize the correlation strength between two strain and tidal time series at different time lags. By identifying the global extreme points of the cross-correlation function, systematically analyze the correlation at different lag times, and then determine the optimal time lag between tunnel structure strain and tide based on the maximum correlation coefficient criterion.

[0017]

[0018] in, To account for the lag time between the tides and the response time, These are the standard deviations of strain and tidal data, respectively. For strain data time series, This is a time series of tidal data. For the mean strain, The average value of the tidal data. For the summation index, Lagging Correlation coefficient over time;

[0019] Step 25: Based on the tidal effect present in the strain data separated by the optimal time delay, obtain the residual strain with the spatial response characteristics of the fracture zone:

[0020]

[0021] in, For total strain; The strain component caused by tidal forces; The residual strain after removing the tidal component;

[0022] Step 3: Constructing the strain mean difference:

[0023] Step 3: Derive the non-uniform deformation mechanism between the tunnel across the fault zone and the adjacent non-fault zone region based on the governing equations of the Euler-Bernoulli beam model.

[0024] Step 32: Use a Gaussian mixture model to perform cluster analysis on the strain data after separating the tidal residual strain, and extract the cluster mean of strain in the fracture zone and non-fracture zone.

[0025] Step 33: Based on the clustering results, clearly define the strain mean of the two key subsets and construct the mean difference;

[0026] Steps three and four: When the tunnel structure undergoes non-uniform deformation, the mean strain value in this stochastic process experiences a structural abrupt change, transforming into a sequential change point detection problem. To further confirm the detection results, a sliding window least squares fitting method is used to calculate the instantaneous slope of the strain difference sequence. To verify, when When the value remains significantly greater than zero, it confirms that uneven deformation is developing;

[0027] Step 4: Identification and precise positioning of uneven deformation of submarine tunnels crossing fault zones:

[0028] Step 4: 1. To address the drift caused by the difference in cluster means between fault zones and non-fault zones, optimize the key hyperparameters in the RF model using the PSO algorithm to build a high-performance classifier.

[0029] Step 42: Based on the identification model in Step 41, effectively determine whether the tunnel has undergone uneven deformation, and use the strain time history curve to preliminarily determine the fracture zone area and the time period in which the deformation occurred, thereby narrowing the analysis scope to a specific time and space interval.

[0030] Step 43: Introduce Local Dynamic Spatio-Temporal Attention (LDSA) to generate a high-resolution three-dimensional feature heatmap for a specific spatiotemporal interval, thereby achieving accurate localization of abnormal regions.

[0031] Compared with the prior art, the present invention has the following advantages:

[0032] This invention employs BOTDA distributed fiber optic sensing to acquire continuous and dense strain data on the tunnel cross-section. Based on the massive amount of distributed strain data combined with the complex geological environment of the cross-fault zone submarine tunnel, it can provide solid data support for the identification, assessment, and handling of uneven deformation of the tunnel structure, thereby enabling the formulation of more scientific management and maintenance decisions and effectively ensuring the safety of subway operation. Attached Figure Description

[0033] Figure 1 A flowchart illustrating the method for identifying and locating uneven deformation in submarine tunnel structures spanning fault zones.

[0034] Figure 2 This refers to the tunnel strain data collected at a certain moment.

[0035] Figure 3 This is the strain data after outlier processing.

[0036] Figure 4 The strain data are denoised using the SG algorithm.

[0037] Figure 5 This is the time history curve of strain data for a certain measuring point over 15 days (the built-in graph is the 48-hour time history curve).

[0038] Figure 6 15-day time history curves of tidal sea level and tidal-strain data.

[0039] Figure 7 The correlation between tidal-strain data before time delay compensation is performed.

[0040] Figure 8 The correlation between the tidal-strain data after time delay compensation is shown.

[0041] Figure 9 This is a spatial curve of strain data with obvious fault zone geological characteristics.

[0042] Figure 10 This is a Gaussian mixture model clustering diagram and probability density distribution diagram of strain data for a certain fault zone and adjacent non-fault zones.

[0043] Figure 11 The time history curves are for the cluster centers (mean strain) of the fracture zone and adjacent non-fracture zones.

[0044] Figure 12 The time history curve represents the difference in the mean strain between the fracture zone and the adjacent non-fracture zone.

[0045] Figure 13 Generate high-resolution 3D feature heatmaps for local dynamic spatiotemporal attention mechanism algorithms. Detailed Implementation

[0046] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0047] This invention provides a method for identifying and locating non-uniform deformation in submarine tunnel structures spanning fault zones. First, strain data from the BOTDA (Board Over-the-Air) model of the submarine tunnel spanning the fault zone is collected. The strain data undergoes outlier processing, fitting, and tidal residual strain separation to obtain strain data with spatial characteristics of the fault zone. Second, based on a Gaussian mixture model, strain data from the fault zone and adjacent non-fault zones are clustered, and the clustered strain mean is extracted to construct a strain mean difference feature vector. The PSO (Plan-Solution-Oriented Strain) algorithm is used to optimize the RF (Radio Frequency) model to identify the time-varying curve of the strain mean difference. For areas experiencing non-uniform deformation, the LDSA (Low-Low-Strain Algorithm) algorithm is used to generate a high-resolution three-dimensional feature heatmap, thereby accurately locating the non-uniform deformation area. Figure 1 As shown, the specific steps are as follows:

[0048] Step 1: Preprocessing of strain data for the submarine tunnel across the fault zone: BOTDA distributed optical fiber was deployed in the subway submarine tunnel to collect strain data. Outlier removal and data fitting were performed on the collected strain data using quartiles and regression methods respectively. Savzky-Golay was then used to reduce noise in the outlier-removed strain data, thus completing the strain data preprocessing. The specific steps are as follows:

[0049] Step 11: Deploy BOTD distributed optical fibers within the subway undersea tunnel to collect BOTDA strain signals. Perform data preprocessing on the collected vibration signals, including outlier handling and normalization. Use regression methods to fit missing values ​​in the strain data, where:

[0050] The specific steps for handling outliers are as follows:

[0051] (1) Select the threshold for outlier identification by calculating quartiles, sort the vibration signals according to amplitude, and calculate the first quartile. and the third quartile ;

[0052] (2) Calculate the threshold for identifying outliers :

[0053]

[0054]

[0055] Will exceed The vibration signal was marked as an anomaly;

[0056] (3) Delete outlier points:

[0057] The normalization process employs a centralized and standardized approach, with the following specific steps:

[0058] (1) Centering is the process of subtracting the average value from the vibration signal after outlier processing, using the following formula:

[0059]

[0060]

[0061] In the formula, Indicates the first [number] after outlier handling A vibration signal, This represents the average value of the vibration signal after outlier processing. This indicates the number of sampling points for the vibration signal after outlier processing. Represents the first after centralized processing One vibration signal;

[0062] (2) The standardization process uses the following formula:

[0063]

[0064] In the formula, Represents the standardized version of the first... A vibration signal, It represents the standard deviation of the vibration signal.

[0065] The missing values ​​of the strain data were fitted using regression analysis, as shown in the following formula:

[0066]

[0067] in, It means should be in The value captured at the missing value, Indicates in The fitted value calculated at the missing values. These represent the coefficients of the regression model. Indicates the regression error. Indicates the spatial location neighborhood The signal value.

[0068] Steps 1 and 2: The Savzky-Golay (SG) algorithm is used to denoise the strain data processed in Step 1. The specific steps are as follows:

[0069] Let a set of data in a window be , , The value is A series of consecutive integer values, To represent the range of data, we now construct a... A polynomial of order 1 fits this set of data:

[0070]

[0071] In the formula, Let the order be the order of the polynomial. , For the first The coefficients of the data, and their values ​​must be less than Once determined Polynomial order and the data to be fitted Then, the central estimate of the data set can be obtained by using the property that the partial derivative is 0. By repeatedly moving the window, the central estimate of other points can be obtained.

[0072] The sum of squared residuals between the fitted data points and the original data points is:

[0073]

[0074] In the formula, To determine the order of the fitted polynomial, To fit the data and determine the flexibility of curve fitting, the least squares method is used. To achieve the best fitting result, the sum of squared residuals should be minimized, i.e. coefficients of the polynomial The partial derivative should be 0:

[0075]

[0076] When the number of single-sided points to be fitted order of polynomial and the data to be fitted Once the information is known, the polynomial can be calculated. The fitted polynomial is used to obtain the estimated value of the center point within the window, and for subsequent points, the operation can be repeated by continuously moving the window.

[0077] Step 2: Tidal Residual Strain Separation: Extract ocean tidal data within the time period of strain data acquisition, decompose the tidal data based on harmonic analysis, and separate the main tidal components under tidal load using least squares spectral analysis. Quantitatively analyze the correlation and time delay effect between tunnel strain and tidal level using Pearson correlation analysis and cross-correlation function methods. Separate the tidal residual strain based on a time delay-corrected tidal strain prediction model to obtain strain with fracture zone spatial response characteristics.

[0078] Step 21: Extract ocean tidal data for the same time period in the target area and decompose the tidal data to extract the amplitude and phase parameters of the main astronomical tides. The process is as follows:

[0079] Tidal height It can be represented as a linear superposition of multiple astronomical tidal constituents:

[0080]

[0081] in, The number of tidal constituents refers to the total number of astronomical tidal constituents included in the analysis. Mean sea level; For tide division The average amplitude; For tide division angular frequency; For tide division The phase delay angle; For astronomical parameter nodal factors; It is the astronomical phase angle; This is the residual term.

[0082] Step 22: Based on least squares spectral analysis, separate... , , , The main tidal components under tidal loads have the following objective function:

[0083]

[0084] in, It is the main lunar semi-diurnal tide, the primary tidal period generated by the moon's gravity, with a period of approximately 12.42 hours; It is the main solar semi-diurnal tide, the primary tidal wave generated by the sun's gravity, with a period of 12.00 hours; It is a solar tide, which is produced by the combined daily cycles of the moon and the sun, with a period of 23.9 hours. It is the primary lunar diurnal tide, a diurnal tide caused by the moon's gravity, with a period of approximately 25.82 hours.

[0085] amplitude and phase With coefficient , The relationship is:

[0086] ,

[0087] Steps 2 and 3: The correlation between tunnel strain and tidal level is quantitatively analyzed using Pearson correlation analysis. To eliminate the influence of dimensions and facilitate comparative analysis, the Z-score normalization method is used.

[0088]

[0089] in, For standardized data values, The original data values, The mean of the data. The standard deviation is 1. The standardized data have a mean of 0 and a standard deviation of 1.

[0090] Using Pearson correlation coefficient Quantifying the degree of linear correlation between tides and strain variables:

[0091]

[0092] in, Data capacity, i.e., the number of data points; The index of the data points, from 1 to ; , They are the first The data includes observations of strain variables and tidal variables; This represents the sample mean of the strain variable; This represents the sample mean of the tidal variable; The value range is [-1, 1], and the closer the absolute value is to 1, the stronger the linear correlation.

[0093] Step 24: Using the cross-correlation function method, quantitatively characterize the correlation strength between two strain and tidal time series at different time lags. By identifying the global extreme points of the cross-correlation function, systematically analyze the correlation at different lag times, and then determine the optimal time lag between tunnel structure strain and tide based on the maximum correlation coefficient criterion.

[0094]

[0095] in, To account for the lag time between the tides and the response time, These are the standard deviations of strain and tidal data, respectively. For strain data time series, This is a time series of tidal data. For the mean strain, The average value of the tidal data. For the summation index, Lagging Correlation coefficient over time.

[0096] Step 25: Based on the tidal effect present in the optimal time-delay separated strain data, obtain the residual strain with the spatial response characteristics of the fracture zone, thereby improving the spatial response characteristics of the fracture zone. The time-delay corrected tidal strain prediction model is as follows:

[0097]

[0098] in, The strain response coefficient; For the lag time between the response and the tide; The reference strain value; This is a time series of tidal water levels.

[0099] The residual strain of the tunnel structure is:

[0100]

[0101] in, For total strain; The strain component caused by tidal forces; The residual strain is the strain after removing the tidal component. The residual strain includes the strain caused by fault zone activity, temperature effects, noise, etc.

[0102] Step 3: Construction of Strain Mean Difference: Based on the governing equations of the Euler-Bernoulli beam model, the non-uniform deformation mechanism between the fracture zone and the adjacent non-fracture zone is derived. A Gaussian mixture model is used to perform cluster analysis on the strain data processed in Steps 1 and 2, extracting the cluster mean strain of the fracture zone and the non-fracture zone. The strain mean difference and time-varying curves between the fracture zone and the non-fracture zone are constructed, and the instantaneous slope of the strain mean difference sequence is extracted. The specific steps are as follows:

[0103] Step 3.1. Based on the governing equations of the Euler-Bernoulli beam model, derive the non-uniform deformation mechanism between the tunnel across the fault zone and the adjacent non-fault zone region. The governing equations are as follows:

[0104]

[0105] In the formula, For the bending stiffness of the lining ring. For vertical settlement of the tunnel, This is the subgrade coefficient. The load is a uniformly distributed external load.

[0106] To solve the above equation, consider its homogeneous form. At the same time, a key parameter is introduced—feature length. :

[0107]

[0108] Feature length Integrating structural stiffness With foundation stiffness is the core dimensionless number for determining the mechanical behavior of a beam-foundation system, thus the governing equation can be written as:

[0109]

[0110] Its general solution is in the form of:

[0111]

[0112] in, , , , The integral constant is determined by the boundary conditions.

[0113] When the tunnel is from a high subgrade coefficient Non-fault zone regions enter low subgrade coefficient When a fracture zone is formed, internal forces will concentrate at the boundary to coordinate deformation, and the moment concentration factor at the interface will increase. It can be characterized as:

[0114]

[0115] in, The reference bending moment under uniform foundation, The maximum bending moment at structural failure, exponential The value is positive. This formula indicates that when the rock mass originates from a stable, non-fractured zone... To the fracture zone ( When this occurs, a huge bending moment will be generated at the junction. With shear force This can lead to huge bending and shear deformations, resulting in significant uneven deformation and seriously threatening the normal operation of tunnels crossing fault zones.

[0116] Step 3.2: Perform cluster analysis on the strain data after separating the tidal residual strain using a Gaussian mixture model, and extract the cluster mean of strain in the fracture zone and non-fracture zone. The deformation data sample matrix in the Gaussian mixture model... probability density The estimated value can be calculated using the formula:

[0117]

[0118] In the formula, To obtain the cumulative sum; The total number of categories in a Gaussian mixture distribution; It is a class of Gaussian mixture distribution. ; For the first The mixing ratio of classes, and ; For the first The mean of a Gaussian-like distribution; For the first Covariance of a Gaussian-like distribution; For the first The probability density function of a Gaussian-like distribution is expressed as follows:

[0119]

[0120] In the formula, For data dimensions, Find the norm of the matrix; For the natural constant An exponential function with base 0; Find the inverse of the matrix.

[0121] If the data sample matrix If each column is independent, then monitoring data can be used to analyze the parameters. , and The maximum likelihood method is used for estimation of the likelihood function. Defined as follows:

[0122]

[0123] In the formula, For the natural constant A logarithmic function with base 0;

[0124] The defined optimization problem is solved using the expectation maximization (EM) algorithm.

[0125] Step 33: Based on the clustering results, the strain mean of the two key subsets can be clearly defined, and the mean difference can be constructed:

[0126] set up This represents the set of data points indicating the identified fault zone region. Represents the set of stable regions outside of fault zones, then The average strain of the two regions at time t can be calculated as follows:

[0127]

[0128] thus, They are respectively fault zones and non-fault zones. , These are sets of strain data points for fracture zones and non-fracture zones, respectively. , Representing the strain data of the fracture zone and the non-fracture zone at time t, respectively, a core observable for identifying non-uniform deformation is constructed: the difference in the mean strain between the fracture zone and the non-fracture zone. :

[0129]

[0130] This physical quantity isolates the common trends brought about by overall load changes, amplifies the differences in relative deformation between different geological units, and provides a direct and sensitive data basis for identifying structural non-uniform deformation caused by adverse geology.

[0131] Under healthy conditions, the mean strain difference between the fracture zone and the non-fracture zone regions It can be viewed as a stationary stochastic process, a model with deterministic constant terms and random noise:

[0132]

[0133] in, The reference term representing stable cooperative deformation, It is a stationary noise process with zero mean and known covariance structure, reflecting normal fluctuations caused by operating loads, measurement errors, etc.

[0134] This random process The covariance structure can be further written as an autoregressive moving average process:

[0135]

[0136] It is independent and identically distributed Gaussian white noise; , These are the autoregressive and moving average coefficients, respectively. This represents the model order.

[0137] Steps three and four: When the tunnel structure undergoes non-uniform deformation, the mean strain value in this random process experiences a structural abrupt change, transforming into a sequential change point detection problem:

[0138] Assuming the tunnel structure remains stable throughout the observation period under healthy conditions:

[0139]

[0140] for The strain mean difference sequence at time points, The total observation time is used when the structure undergoes uneven deformation and there are unknown change points. The process involves drift:

[0141]

[0142] in, For drift rate, The point of change (the moment when the structural state changes). Drift rate, characterizing the rate of development of non-uniform deformation, when This indicates that the uneven deformation continues to worsen.

[0143] To further confirm the test results, the instantaneous slope of the strain difference sequence was calculated using sliding window least squares fitting for verification.

[0144]

[0145] in, for The instantaneous slope estimate at time t. For the width of the sliding window, This is the time index within the window. When... When the value remains significantly greater than zero, it confirms that uneven deformation is developing.

[0146] Step 4: Identification and Precise Location of Non-uniform Deformation of Submarine Tunnels Across Fault Zones: A high-performance classifier based on the PSO algorithm and optimized RF model is constructed to identify the regions and times of non-uniform deformation in the time-varying curve of strain mean difference. A local dynamic spatiotemporal attention mechanism algorithm is used to generate a high-resolution three-dimensional feature heatmap for precise location, thus realizing the identification and precise location of non-uniform deformation of submarine tunnels across fault zones. The specific steps are as follows:

[0147] Step 4.1: To address the drift caused by the difference in cluster means between fault zones and non-fault zones, the PSO algorithm is used to optimize the key hyperparameters in the RF model to build a high-performance classifier. Random Forest (RF) is an ensemble learning algorithm that classifies data by constructing multiple decision trees and combining their results. The RF classification function is:

[0148]

[0149] In the formula, The number of decision trees affects model complexity and computational cost; This is the maximum depth of the tree, controlling model complexity and the risk of overfitting; This is the minimum number of samples required for internal node splitting, used as the stopping condition for tree growth; This is the minimum number of samples required for a leaf node to prevent overfitting. The number of features considered when searching for the optimal split affects the diversity and generalization ability of the tree.

[0150] Particle Swarm Optimization (PSO) is a global optimization algorithm characterized by fast convergence and resistance to getting trapped in local optima. PSO iterates continuously until it finds the optimal solution. The update formulas for particle velocity and position are shown below:

[0151]

[0152]

[0153] In the formula, For the first The particle in the first The position vector at the next iteration. For the first The particle in the first The velocity vector at the next iteration This represents the number of iterations. Inertial weights; and The acceleration factor is the learning factor for both the "individual" and "global" aspects. and A random number between 0 and 1; and Particles In dimensions The individual optimal solution and the global optimal solution.

[0154] Step 42: Based on the identification model in Step 41, it can effectively determine whether the tunnel has undergone uneven deformation, and use the strain time history curve to preliminarily determine the fracture zone area and the time period of deformation, thereby narrowing the analysis scope to a specific spatiotemporal interval, which can significantly reduce the calculation time for subsequent accurate positioning.

[0155] Step 43: Based on this, a Local Dynamic Spatio-Temporal Attention (LDSA) mechanism is introduced to generate a high-resolution three-dimensional feature heatmap for a specific spatiotemporal interval, thereby achieving accurate localization of abnormal regions.

[0156] The LDSA mechanism aims to capture the local dependencies and dynamic patterns of strain data in both spatial and temporal dimensions. For anomalous time periods... and abnormal spatial location strain characteristics at the location Its attention weight It can be represented as:

[0157]

[0158] in:

[0159]

[0160] In the formula, For position In time The normalized attention weights represent the importance of the spatiotemporal location to the current prediction. For spatial neighborhood set; A set of time windows; This is the spatial attention weight matrix; This is the time attention weight matrix; This is the spatiotemporal interaction weight matrix; Output a weight vector for the attention function; This is the attention bias vector; For spatial feature vectors; This is a time feature vector; This is the vector concatenation operator.

[0161] Example:

[0162] The BOTDA distributed optical fiber is used in the Qingdao Metro Line 8 undersea tunnel section from Dayang Station to Qingdao North Station. The tunnel crosses three major fault zones. The total length of the optical cable is 7.8km along the entire track. The optical cable is a 3mm outer diameter tight-buffered optical cable. The distributed optical fiber acoustic wave detector (DAS) equipment is used as the light source emission device and data acquisition device.

[0163] Automatically acquire strain signal data from BOTDA; the acquired strain signal along the tunnel space at a certain moment is as follows: Figure 2 As shown, due to the characteristics of the BOTDA system, the directly acquired strain signals generally have relatively low signal quality. Therefore, the strain data preprocessing method proposed in step one is first used to preprocess the acquired BOTDA strain data to remove outliers. The outlier processing results are shown below. Figure 3 As shown, the SG signal denoising algorithm is used to denoise the strain data. The denoised data is as follows: Figure 4 As shown.

[0164] Based on steps 21 to 23, which use the Pearson correlation coefficient between strain and tidal data as an indicator, this study analyzes the correlation between the 15-day time series of strain and tidal data at different time lags using cross-correlation functions. Correlation analysis is performed on the time lags in both fault zone and non-fault zone regions. The results are as follows: Figure 6 As shown, the overall time delay in the fault zone region is approximately 4.2 hours, while the overall time delay in the non-fault zone region is approximately 6.5 hours.

[0165] By compensating for the time lag between strain data from different measuring points and the tidal effect, the correlation between tidal and strain data in terms of time characteristics is improved. Before compensation, the Pearson correlation coefficient between the two was only in the range of 0.1 to 0.15. After time lag compensation, the correlation coefficient increased to 0.61 to 0.80. Compensation was performed on tidal-strain data at different locations along the tunnel space. The correlations between tidal-strain data before and after time lag compensation are as follows: Figure 7 , Figure 8 As shown, strain with fracture zone spatial response characteristics can be obtained by removing tidal residual strain.

[0166] Based on the analysis of tidal residual strain in steps two to two, and after strain data preprocessing and tidal residual strain separation in steps one and two, strain data with obvious fault zone geological characteristics are obtained. The strain data spatial curves are shown below. Figure 9 As shown.

[0167] Based on the strain mean difference construction method proposed in step three, three representative fault influence zones (F3, F4, F5) were selected. The range of the tunnel section across the fault zone was determined by the previous geological radar exploration data and the spatial distribution characteristics of BOTDA fiber optic measuring points. Strain data of the fault zone section and the adjacent non-fault zone section were extracted respectively.

[0168] The Gaussian mixture model proposed in step 31 is used to cluster the interval strain data. Each selected interval strain data includes the strain data of the fracture zone and the strain data of the adjacent left or right non-fracture zone. Based on the Gaussian mixture model, cluster analysis is performed on the strain data of a certain fracture zone and the adjacent non-fracture zone. The clustering results are as follows: Figure 10 As shown.

[0169] Strain monitoring data for approximately 1050 days between 2022 and 2026 were selected. Since the tunnel is a newly built tunnel and its structure is in good condition, in order to further verify the effectiveness of the proposed method, while keeping the monitoring data of the non-fracture zone unchanged, a 20-meter section in the fault zone was simulated to simulate the uneven deformation caused by external factors (such as changes in water and soil pressure, fault zone displacement, etc.).

[0170] Based on the actual strain variation patterns in the early stages of the fault zone, starting from the 800th day, the data is replaced with simulated strain values ​​that increase at a certain gradient to construct mean difference anomalies. The strain mean time history curves of the constructed fault zone and non-fault zone cluster centers are shown below. Figure 11 As shown.

[0171] Based on the formula derived in step 3, a strain mean difference feature is constructed, and the constructed PSO algorithm is used to optimize the RF model and the high-performance classifier model to detect the strain mean difference time history curve. The detection results are as follows: Figure 12As shown, the results indicate that when the structure undergoes non-uniform deformation, the time history curve of the strain mean difference has a significant response. Based on this time history curve, the fracture zone and approximate time of the non-uniform deformation can be determined, thereby greatly reducing the target to be detected.

[0172] Based on the local dynamic spatiotemporal attention mechanism algorithm proposed in step four, a high-resolution three-dimensional feature heatmap is drawn for this region and this time period, such as... Figure 13 As shown, the uneven deformation that occurred in this area can be clearly seen.

Claims

1. A method for identifying and locating non-uniform deformation in a submarine tunnel structure spanning a fault zone, characterized in that... The method includes the following steps: Step 1: Preprocessing of strain data for submarine tunnels crossing fault zones: Step 11: Deploy BOTDA distributed optical fiber in the subway undersea tunnel, collect BOTDA strain signals, and perform outlier processing and normalization on the collected vibration signals. Use regression method to fit the missing values ​​of the strain data. Step 1 and Step 2: The Savzky-Golay (SG) algorithm is used to reduce noise in the strain data processed in Step 1, thus realizing the preprocessing of the strain data; Step 2: Tidal residual strain separation: Step 21: Extract ocean tidal data for the same time period in the target area and decompose the tidal data to extract the amplitude and phase parameters of the main astronomical tides: in, The number of tidal constituents refers to the total number of astronomical tidal constituents included in the analysis. Tidal height; Mean sea level; For tide division The average amplitude; For tide division angular frequency; For tide division The phase delay angle; For astronomical parameter nodal factors; It is the astronomical phase angle; For residual terms; Step 22: Based on least squares spectral analysis, the main tidal components under tidal load are separated. The objective function is: in, , For coefficients; Steps 2 and 3: Quantitatively analyze the correlation between tunnel strain and tidal level using Pearson correlation analysis. Step 24: Using the cross-correlation function method, quantitatively characterize the correlation strength between two strain and tidal time series at different time lags. By identifying the global extreme points of the cross-correlation function, systematically analyze the correlation at different lag times, and then determine the optimal time lag between tunnel structure strain and tide based on the maximum correlation coefficient criterion. in, To account for the lag time between the tides and the response time, These are the standard deviations of strain and tidal data, respectively. For strain data time series, This is a time series of tidal data. For the mean strain, The average value of the tidal data. For the summation index, Lagging Correlation coefficient over time; Step 25: Based on the tidal effect present in the strain data separated by the optimal time delay, obtain the residual strain with the spatial response characteristics of the fracture zone: in, For total strain; The strain component caused by tidal forces; The residual strain after removing the tidal component; Step 3: Constructing the strain mean difference: Step 3: Derive the non-uniform deformation mechanism between the tunnel across the fault zone and the adjacent non-fault zone region based on the governing equations of the Euler-Bernoulli beam model. Step 32: Use a Gaussian mixture model to perform cluster analysis on the strain data after separating the tidal residual strain, and extract the cluster mean of strain in the fracture zone and non-fracture zone. Step 33: Based on the clustering results, clearly define the strain mean of the two key subsets and construct the mean difference; Steps three and four: When the tunnel structure undergoes non-uniform deformation, the mean strain value in this stochastic process experiences a structural abrupt change, transforming into a sequential change point detection problem. To further confirm the detection results, a sliding window least squares fitting method is used to calculate the instantaneous slope of the strain difference sequence. To verify, when When the value remains significantly greater than zero, it confirms that uneven deformation is developing; Step 4: Identification and precise positioning of uneven deformation of submarine tunnels crossing fault zones: Step 4:

1. To address the drift caused by the difference in cluster means between fault zones and non-fault zones, optimize the key hyperparameters in the RF model using the PSO algorithm to build a high-performance classifier. Step 42: Based on the identification model in Step 41, effectively determine whether the tunnel has undergone uneven deformation, and use the strain time history curve to preliminarily determine the fracture zone area and the time period in which the deformation occurred, thereby narrowing the analysis scope to a specific time and space interval. Step 43: Introduce a local dynamic spatiotemporal attention mechanism to generate a high-resolution three-dimensional feature heatmap for a specific spatiotemporal interval, thereby achieving accurate localization of abnormal areas.

2. The method for identifying and locating uneven deformation of a cross-fault zone submarine tunnel structure according to claim 1, characterized in that... The specific steps of step one are as follows: Step 11: Deploy BOTDA distributed optical fiber in the subway undersea tunnel, collect BOTDA strain signals, and perform outlier processing and normalization on the collected vibration signals. Use regression method to fit the missing values ​​of the strain data. Step 1 and Step 2: The SG algorithm is used to reduce noise in the strain data processed in Step 1, thus realizing the preprocessing of the strain data.

3. The method for identifying and locating uneven deformation of a cross-fault zone submarine tunnel structure according to claim 2, characterized in that... The specific steps for handling outliers in step one are as follows: (1) Select the threshold for outlier identification by calculating quartiles, sort the vibration signals according to amplitude, and calculate the first quartile. and the third quartile ; (2) Calculate the threshold for identifying outliers : Will exceed The vibration signal was marked as an anomaly; (3) Delete outlier points: The normalization process employs a centralized and standardized approach, with the following specific steps: (1) Centralized processing: In the formula, Indicates the first [number] after outlier handling A vibration signal, This represents the average value of the vibration signal after outlier processing. This indicates the number of sampling points for the vibration signal after outlier processing. Indicates the first after centralized processing One vibration signal; (2) Standardization process: In the formula, Represents the standardized version of the first... A vibration signal, This represents the standard deviation of the vibration signal; The formula for fitting missing values ​​in strain data using regression is expressed as follows: in, It means should be in The value captured at the missing value, Indicates in The fitted value calculated at the missing values. These represent the coefficients of the regression model. Indicates the regression error. Indicates the spatial location neighborhood The signal value.

4. The method for identifying and locating uneven deformation of a trans-fault zone submarine tunnel structure according to claim 1, characterized in that... In step two, the amplitude and phase With coefficient , The relationship is: ， 。 5. The method for identifying and locating uneven deformation of a cross-fault zone submarine tunnel structure according to claim 1, characterized in that... In steps two and three, the Pearson correlation coefficient is used. Quantifying the degree of linear correlation between tides and strain variables: in, Data capacity, i.e., the number of data points; For indexes of data points; , They are the first The data includes observations of strain variables and tidal variables; The sample mean of the strain variable; This represents the sample mean of the tidal variable; The value range is [-1, 1].

6. The method for identifying and locating non-uniform deformation of a trans-fault zone submarine tunnel structure according to claim 1, characterized in that... In step two five, , The strain response coefficient; The lag time between strain and tides; The reference strain value; This is a time series of tidal water levels.

7. The method for identifying and locating non-uniform deformation of a trans-fault zone submarine tunnel structure according to claim 1, characterized in that... In step 31, the governing equations are as follows: In the formula, For the bending stiffness of the lining ring. For vertical settlement of the tunnel, This is the subgrade coefficient. For uniformly distributed external loads; To solve the above equation, consider its homogeneous form. At the same time, feature length is introduced. : The governing equations are written as follows: Its general solution is in the form of: in, , , , The integral constant is determined by the boundary conditions; When the tunnel is from a high subgrade coefficient Non-fault zone regions enter low subgrade coefficient When a fracture zone is formed, internal forces will concentrate at the boundary to coordinate deformation, and the moment concentration factor at the interface will increase. Characterized as: in, The reference bending moment under uniform foundation, The maximum bending moment at structural failure, exponential It is a positive number.

8. The method for identifying and locating non-uniform deformation of a trans-fault zone submarine tunnel structure according to claim 1, characterized in that... In step three, let This represents the set of data points indicating the identified fault zone region. Represents the set of stable regions without fault zones, then The average strain of the two regions at time t is calculated as follows: in, They are respectively fault zones and non-fault zones. , These are sets of strain data points for fracture zones and non-fracture zones, respectively. , Representing fault zones and non-fault zones respectively. Strain data at any given moment; Difference in mean strain between fracture zone and non-fracture zone regions for: In a healthy state, ,in, The reference term representing stable cooperative deformation, It is a stationary noise process with zero mean and a known covariance structure. This stochastic process The covariance structure can be further written as an autoregressive moving average process: in, It is independent and identically distributed Gaussian white noise; , These are the autoregressive and moving average coefficients, respectively. This represents the model order.

9. The method for identifying and locating non-uniform deformation of a trans-fault zone submarine tunnel structure according to claim 1, characterized in that... In steps three and four, it is assumed that the tunnel structure remains stable throughout the observation period under healthy conditions: for The strain mean difference sequence at time t, The total observation time is used when the structure undergoes uneven deformation and there are unknown change points. The process involves drift: in, For drift rate, To change the position of the point, Drift rate; The instantaneous slope of the strain difference sequence was calculated using sliding window least squares fitting for verification. in, for The instantaneous slope estimate at time t. For the width of the sliding window, This is the time index within the window.

10. The method for identifying and locating non-uniform deformation of a trans-fault zone submarine tunnel structure according to claim 1, characterized in that... In step four, the abnormal time period is... and abnormal spatial location strain characteristics at the location Its attention weight Represented as: in: In the formula, For position In time The normalized attention weights represent the importance of the spatiotemporal location to the current prediction. For spatial neighborhood set; A set of time windows; This is the spatial attention weight matrix; This is the time attention weight matrix; This is the spatiotemporal interaction weight matrix; Output a weight vector for the attention function; This is the attention bias vector; For spatial feature vectors; This is a time feature vector; This is the vector concatenation operator.

Images