Subway tunnel segment structure monitoring method, system, equipment and medium

Abstract

The invention provides a subway tunnel segment structure monitoring method, system and device and a medium, and belongs to the technical field of subway tunnel monitoring, and the method comprises the steps: obtaining a time sequence signal comprising at least two monitoring variables; the time sequence signal is collected by an optical fiber sensor array laid on the inner wall of the subway tunnel; constructing a correlation matrix among different monitoring variables at the current moment based on the time sequence signal; the correlation matrix is used for representing a cooperative relationship between different monitoring variables; based on a pre-trained time convolutional network model, obtaining a corresponding anomaly score according to the time sequence signal; and according to the correlation matrix and the abnormal score, combining and analyzing whether the segment structure is abnormal or not. The duct piece structure change trend is accurately judged, and subway operation safety is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of subway tunnel monitoring technology, specifically to a method, system, equipment, and medium for monitoring the structure of subway tunnel segments. Background Technology

[0002] With the rapid expansion of urban rail transit networks and the continuous increase in operating mileage, the long-term operational safety of subway tunnels has increasingly become a core concern in the field of public safety.

[0003] As the main load-bearing components of the tunnel structure, the structural health of subway tunnel segments directly determines the stability and safety of the entire tunnel. During long-term operation, the segment structure is affected by various factors such as surrounding soil pressure, groundwater erosion, cyclic dynamic loads from trains, and nearby construction, which may lead to the occurrence and development of defects such as cracks, leakage, convergence deformation, and bolt loosening. Therefore, efficient, accurate, and long-term monitoring of the segment structure and timely prediction of potential defects are of paramount importance for preventing major safety accidents, optimizing maintenance strategies, and ensuring the safe operation of subway lines.

[0004] Currently, traditional monitoring methods for subway tunnel segments have many limitations, such as vibration fiber optic sensors, resistive fiber optic sensors, steel wire fiber optic sensors, and piezoelectric point-type fiber optic sensors. These traditional subway tunnel segment sensors generally suffer from drawbacks in practical applications, including difficulty in deployment, poor anti-interference capabilities, poor corrosion resistance, susceptibility to damage, and data distortion. Existing monitoring methods often only acquire data for limited time periods and localized areas, making it difficult to effectively analyze and accurately predict the long-term changing trends of the segment structure. Summary of the Invention

[0005] In view of this, it is necessary to provide a method, system, equipment and medium for monitoring the structure of subway tunnel segments, so as to solve the technical problem that the existing technology cannot effectively monitor the structure of subway tunnel segments in a comprehensive, long-term and accurate manner.

[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a method for monitoring the structure of subway tunnel segments, comprising: Acquire time-series signals including at least two monitored variables; the time-series signals are collected by an array of fiber optic sensors laid on the inner wall of the subway tunnel; Based on the time series signal, a correlation matrix is ​​constructed between the different monitored variables at the current time; the correlation matrix is ​​used to characterize the cooperative relationship between the different monitored variables. Based on a pre-trained temporal convolutional network model, the corresponding anomaly score is obtained according to the time series signal; Based on the correlation matrix and the anomaly score, an analysis is conducted to determine whether any anomalies have occurred in the segment structure.

[0007] In one possible implementation, acquiring the time-series signal comprising at least two monitored variables includes: Acquire the physical signal collected by each fiber optic sensor in the fiber optic sensor array; the physical signal includes the wavelength parameter of the fiber optic sensor; the wavelength parameter corresponds to the monitored variable; The wavelength parameters of the multiple physical signals are demodulated into the time-series signal according to the time order and the number of the fiber optic sensor.

[0008] In one possible implementation, constructing a correlation matrix between different monitored variables at the current time based on the time series signal includes: Extract all data of each monitored variable within the same target time window to obtain the corresponding time segment; the target time window is represented as [tw, t], where t is the current time and w is the preset window length; Calculate the average correlation strength of any two different monitored variables corresponding to the time series segments; The average correlation strength corresponding to each of the two different monitored variables is filled into an n×n matrix to obtain the correlation matrix corresponding to the current time, where n is a positive integer greater than 2.

[0009] In one possible implementation, calculating the average correlation strength of the time series corresponding to any two different monitored variables includes: The absolute correlation strength of any two different monitored variables corresponding to the time series segments is calculated: ; or, The relative correlation strength of any two different monitored variables corresponding to the time series segments is calculated: ; in, The absolute correlation strength between i and j Let represent the relative correlation strength between i and j, where i and j are two distinct monitored variables, and n is the total number of monitored variables. Let w be an integer from 0 to w, and ||·|| be the L2 norm function.

[0010] One possible implementation also includes: Normal feature sequences and abnormal feature sequences are obtained and divided into training sets and validation sets; the normal feature sequences are time series signals collected when the pipe structure is in a healthy state, and the abnormal feature sequences are time series signals collected when the pipe structure is in an unhealthy state. Both the normal feature sequences and the abnormal feature sequences are labeled with the monitoring variables. The temporal convolutional network model is trained based on the normal feature sequences and the abnormal feature sequences in the training set. Based on a multivariate Gaussian distribution, calculate the abnormality score corresponding to each of the normal feature sequence and the abnormal feature sequence in the validation set. Iterate through different candidate thresholds and calculate the accuracy, recall, and harmonic mean of anomaly detection for each candidate threshold; the harmonic mean is calculated based on the accuracy and recall. The highest harmonic mean is determined as the anomaly score threshold.

[0011] In one possible implementation, the temporal convolutional network module, which obtains the corresponding anomaly score based on the time-series signal, includes: The time series signal is input into the temporal convolutional network model to obtain the predicted value of the monitored variable at the next time step. An error vector is calculated based on the predicted and actual measured values ​​of the monitored variables; The anomaly score is calculated based on the error vector using a multivariate Gaussian distribution and Mahalanobis distance algorithm.

[0012] In one possible implementation, the step of analyzing whether the segment structure exhibits anomalies based on the correlation matrix and the anomaly score includes: If the correlation change rate between the monitored variables calculated based on the correlation matrix is ​​greater than or equal to a preset change rate, and the change rate of the abnormal score is greater than or equal to a preset rate, the abnormality type is determined to be a sudden phase abnormality. If the rate of change of the correlation between the monitored variables calculated based on the correlation matrix is ​​less than the preset rate of change, and the rate of change of the abnormal score is less than the preset rate, the abnormality type is determined to be a gradual phase abnormality.

[0013] Secondly, the present invention also provides a subway tunnel segment structure monitoring system, comprising: The acquisition module is used to acquire time series signals including at least two monitoring variables; the time series signals are collected by an array of fiber optic sensors laid on the inner wall of the subway tunnel. The construction module is used to construct a correlation matrix between different monitored variables at the current time based on the time series signal; the correlation matrix is ​​used to characterize the cooperative relationship between different monitored variables. The prediction module is used to obtain the corresponding anomaly score based on the time series signal, using a pre-trained temporal convolutional network model. The generation module is used to analyze whether the segment structure has any abnormalities based on the correlation matrix and the anomaly score.

[0014] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the subway tunnel segment structure monitoring method described in any of the above implementations.

[0015] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in the subway tunnel segment structure monitoring method described in any of the above implementations.

[0016] The beneficial effects of this invention are as follows: The subway tunnel segment structure monitoring method provided by this invention first collects time-series signals of at least two monitoring variables and constructs a correlation matrix to characterize the synergistic relationship between these variables. Compared with single-variable monitoring, constructing a correlation matrix can more comprehensively capture the state information of the subway tunnel segment structure, because anomalies in the subway tunnel segment structure are often reflected in multiple monitoring variables, such as temperature, stress, and strain, which may be interrelated. By considering the synergistic relationship between these variables through the correlation matrix, it is possible to more accurately determine whether anomalies have occurred in the segment structure. Using a pre-trained temporal convolutional network model to obtain corresponding anomaly scores based on the time-series signals, a large number of time-series signals can be processed and anomaly scores obtained in a short time, thereby achieving rapid anomaly detection. This is crucial for scenarios like subway tunnels that require real-time monitoring, enabling timely detection of anomalies in the segment structure so that appropriate measures can be taken to prevent accidents. Furthermore, the analysis of segment structure anomalies is performed based on the correlation matrix and anomaly scores obtained from time series signals. This eliminates the need for manual inspection of each monitoring variable, saving significant manpower and time costs while reducing subjective judgment. Moreover, the cross-validation mechanism combining the correlation matrix and anomaly scores can accurately determine the trend and type of anomaly in the segment structure, reducing the possibility of false alarms and missed alarms, enhancing the reliability of monitoring results, and thus allowing for more rational scheduling of maintenance work, improving the utilization efficiency of maintenance resources, further reducing maintenance costs while ensuring the safe operation of the subway. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of an embodiment of the subway tunnel segment structure monitoring method provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of an embodiment of S100; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S200; Figure 4 For the present invention Figure 1 A flowchart illustrating an embodiment of creating a temporal convolutional network model; Figure 5 For the present invention Figure 1 A schematic diagram of an embodiment of S300; Figure 6 This is a schematic flowchart of an example of the subway tunnel segment structure monitoring method provided by the present invention. Figure 7 This is a schematic diagram of an embodiment of the subway tunnel segment structure monitoring system provided by the present invention; Figure 8 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] Before demonstrating the embodiments, the following terms will be explained.

[0024] TCN (Temporal Convolutional Network) is an architecture based on convolutional neural networks, specifically designed for processing time series data. TCN uses convolution operations to capture local and global dependencies in time series data.

[0025] The F1 score is the harmonic mean of precision and recall, used to evaluate the performance of a classification model: F1 = 2 × [(Precision × Recall) / (Precision + Recall)]. The F2 score is the weighted harmonic mean of precision and recall, with a greater emphasis on recall: F2 = 5 × [(Precision × Recall) / (4 × Precision + Recall)].

[0026] FBG (Fiber Bragg Grating) is a structure that forms a periodic change in refractive index within the core of an optical fiber. The main function of an FBG is to reflect light of a specific wavelength, which is determined by the grating's period and the fiber's effective refractive index.

[0027] This invention provides a method, system, equipment, and medium for monitoring the structure of subway tunnel segments, which are described below.

[0028] Figure 1 This is a schematic flowchart of an embodiment of the subway tunnel segment structure monitoring method provided by the present invention, as shown below. Figure 1 As shown, the monitoring methods for subway tunnel segment structures include: S100. Acquire a time series signal including at least two monitoring variables; the time series signal is collected by an array of fiber optic sensors laid on the inner wall of the subway tunnel. S200. Based on the time series signal, construct a correlation matrix between different monitored variables at the current time; the correlation matrix is ​​used to characterize the cooperative relationship between different monitored variables. S300. Based on a pre-trained temporal convolutional network model, obtain the corresponding anomaly score according to the time series signal; S400. Based on the correlation matrix and the anomaly score, analyze whether the segment structure has any anomalies.

[0029] It should be noted that: along the longitudinal and circumferential directions of the inner wall of the subway tunnel segment, optical cables integrating multiple fiber optic (FBG) sensors are laid at specific intervals (e.g., 1-2 meters) to form a sensing network covering the entire area. Each FBG sensor is a monitoring point. A fiber optic sensor demodulator is used to scan the entire fiber optic sensor array at high frequencies (e.g., kHz levels). The demodulator emits broadband light into the optical fiber and receives the light signal of a specific wavelength (λ) reflected back from each FBG fiber optic sensor, modulated by stress and temperature. The time-series signal obtained in step S100 is input into a pre-trained temporal convolutional network (TCN) model. The TCN model analyzes the time-series signal based on patterns learned from massive amounts of normal data to obtain the output results. Cross-validation is performed based on the correlation matrix and the anomaly score to analyze whether there are anomalies in the subway tunnel segment structure.

[0030] In summary, the subway tunnel segment structure monitoring method provided in this invention first collects time-series signals of at least two monitoring variables and constructs a correlation matrix to characterize the synergistic relationship between these variables. Compared to single-variable monitoring, constructing a correlation matrix can more comprehensively capture the state information of the subway tunnel segment structure, because anomalies in the subway tunnel segment structure often manifest in multiple monitoring variables, such as temperature, stress, and strain, which may be interrelated. Considering the synergistic relationship between these variables through the correlation matrix allows for a more accurate determination of whether anomalies have occurred in the segment structure. Utilizing a pre-trained temporal convolutional network model to obtain corresponding anomaly scores based on the time-series signals enables the processing of a large number of time-series signals and the generation of anomaly scores in a short time, thereby achieving rapid anomaly detection. This is crucial for scenarios like subway tunnels that require real-time monitoring, enabling timely detection of anomalies in the segment structure so that appropriate measures can be taken to prevent accidents. Furthermore, the analysis of segment structure anomalies is performed based on the correlation matrix and anomaly scores obtained from time series signals. This eliminates the need for manual inspection of each monitoring variable, saving significant manpower and time costs while reducing subjective judgment. Moreover, the cross-validation mechanism combining the correlation matrix and anomaly scores can accurately determine the trend and type of anomaly in the segment structure, reducing the possibility of false alarms and missed alarms, enhancing the reliability of monitoring results, and thus allowing for more rational scheduling of maintenance work, improving the utilization efficiency of maintenance resources, further reducing maintenance costs while ensuring the safe operation of the subway.

[0031] In some embodiments of the present invention, such as Figure 2 As shown, step S100 includes: S110. Acquire the physical signal collected by each fiber optic sensor in the fiber optic sensor array; the physical signal includes the wavelength parameter of the fiber optic sensor; the wavelength parameter corresponds to the monitoring variable; S120. The wavelength parameters of the multiple physical signals are demodulated into time-series signals according to the time order and the number of the optical fiber sensor.

[0032] It should be noted that each monitored variable (i.e., each FBG fiber optic sensor) corresponds to a time-ordered sequence of wavelength values, denoted as xi(t). When the tube structure experiences strain (tension or compression) or changes in ambient temperature, the grid period and effective refractive index of the fiber optic sensor array change. This physical change directly causes a linear shift (Δλ) in the center wavelength (λ_B) of the reflected light. The fiber optic sensor demodulator scans the entire fiber optic sensor array at a fixed, high frequency (e.g., 1 kHz, or 1000 times per second). The demodulator emits broadband light into the fiber and rapidly acquires the spectral signal (i.e., physical signal) reflected back from each FBG fiber optic sensor. Through built-in algorithms (such as peak detection algorithms), the demodulator calculates the center wavelength value (i.e., wavelength parameter) reflected by each FBG fiber optic sensor in real time and with high precision. Each fiber optic sensor corresponds to a monitored variable, and its output is a time-ordered sequence of wavelength parameters. When outputting data, the fiber optic sensor demodulator binds each wavelength parameter (λ) to two key metadata elements (timestamp and fiber optic sensor number), aligning and arranging the data streams from all fiber optic sensors in chronological order. This process continues until an n-dimensional time-series signal is obtained.

[0033] In summary, the fiber optic sensor demodulator collects wavelength parameters from multiple measurement areas at corresponding times and demodulates them into a time-series signal, denoted as […]. The fiber optic sensor demodulator captures the center wavelength drift of each grating in real time at high frequencies (kHz level) and demodulates it into a high-fidelity, continuous time-series signal of vibrational displacement (monitored variable j) or slow strain change (monitored variable i). This is a set of digital signals output by the fiber optic sensor demodulator, arranged in chronological order, reflecting the wavelength values ​​of each FBG fiber optic sensor. The correlation matrix for each time point is calculated based on the raw accumulated data. Given an n-dimensional time series , It is an n-dimensional time series matrix. The value of the i-th monitored variable at the t-th time step refers to the original wavelength value measured by the i-th fiber optic sensor at the t-th sampling time.

[0034] In some embodiments of the present invention, such as Figure 3 As shown, step S200 includes: S210. Extract all data of each monitored variable within the same target time window to obtain the corresponding time segment; the target time window is represented as [tw, t], where t is the current time and w is the preset window length; S220. Calculate the average correlation strength of the time series corresponding to any two different monitored variables; S230. The average correlation strength corresponding to each of the two different monitored variables is filled into an n×n matrix to obtain the correlation matrix corresponding to the current time, where n is a positive integer greater than 2.

[0035] It should be noted that the system presets a window length (w). This parameter defines the amount of historical data required for analysis. The unit of w is time steps, i.e., the number of data points. For example, if the data sampling frequency is 1 time / minute, setting w=60 means analyzing data from the past 60 minutes. t represents the current time, which is the baseline time point the system is analyzing. From the stored continuous time series signal, for each monitored variable (i.e., each fiber optic sensor), a segment of data is extracted from time tw to time t. Assuming there are n monitored variables, for the i-th monitored variable, its time segment is denoted as... For the j-th monitored variable, its time series is denoted as . Iterate through all possible time segments corresponding to the monitored variable pairs (i, j), and calculate the average association strength of all monitored variable pairs (i, j) for each time segment, where i and j are the numbers of the monitored variables, and i ≠ j. For example, for 3 monitored variables, we need to calculate 6 pairs: (1, 2), (1, 3), (2, 1), (2, 3), (3, 1), and (3, 2). Create a two-dimensional array (matrix) of size n x n with a length of M. t It initializes all its elements to 0, and fills the matrix M with the average correlation strength of each pair of variables (i, j) calculated in step S220. t By taking the i-th row and j-th column, we can obtain a complete, symmetric correlation matrix (if the above standardization method is used).

[0036] In this embodiment, by constructing a correlation matrix, the monitoring perspective is elevated from isolated points to a network of relationships composed of all fiber optic sensors. The correlation matrix is ​​extremely sensitive to relationship breakdowns and can detect abnormal signs before physical deformation reaches dangerous values, thus achieving true early warning. For example, structural damage (such as microcracks) often only leads to the redistribution of internal forces and the disruption of cooperative relationships in the early stages, rather than immediately causing the absolute deformation at a certain point to exceed the standard.

[0037] In some embodiments of the present invention, calculating the average correlation strength of the time series corresponding to any two different monitored variables includes: The absolute correlation strength of any two different monitored variables corresponding to the time series segments is calculated: ;or, The relative correlation strength of any two different monitored variables corresponding to the time series segments is calculated: ; in, The absolute correlation strength between i and j Let represent the relative correlation strength between i and j, where i and j are two distinct monitored variables, and n is the total number of monitored variables. Let w be an integer from 0 to w, and ||·|| be the L2 norm function.

[0038] It should be noted that: given a time series segment from time tw to time t, where t is the target time for the current analysis and w is the length of the time series window (i.e., the duration of historical data used for analysis, in time steps), the time series segment from tw to time t is obtained by extracting accumulated raw data. Assume that two of these time series segments are... ,and Calculate the correlation between the two sequences to obtain the correlation matrix at time t. The correlation between the two time segments can be either absolute correlation strength or relative correlation strength.

[0039] The formula for calculating the absolute correlation strength is as follows: (Equation 1).

[0040] The formula for calculating the relative correlation strength is as follows: (Equation 2).

[0041] and The essence is to extract the dynamic change process of two monitoring variables within the same time window, where t is the target time of the current analysis, w is the length of the time window (i.e. the duration of historical data used for analysis, in units of time steps), and i and j are the monitoring variable numbers, respectively. Let be the monitoring value of the i-th monitoring variable at time t-δ (δ is an integer from 0 to w, representing δ time steps back). It represents the complete time series of the i-th monitored variable from time tw to t (composed of monitored values ​​of w+1 consecutive time steps).

[0042] Equations (1) and (2) are used to calculate the average correlation strength between the i-th and j-th monitoring variables within the time window (length w) corresponding to time t, providing characteristic basis for subsequent judgment on whether the segment structure is abnormal. Summation term Iterate through each time step within the time window (from time tw to time t, a total of w+1 steps), calculate the product of the monitoring value of the i-th monitoring variable and the monitoring value of the j-th monitoring variable in that step, and then sum all the products. This summation reflects the overall correlation between the two monitoring variables throughout the entire time window (if the two monitoring variables change synchronously, the product will remain positive and large; if they change in opposite directions, the product will be small or even negative).

[0043] Equation 1 / w averages the summation result by dividing it by the time window length w, eliminating differences in the overall correlation strength caused by different window lengths (e.g., the summation results of w=5 and w=10 cannot be directly compared; averaging them can unify them into an average correlation strength per unit time step). Equation 2 takes the norm (usually the L2 norm, i.e., the absolute value) of the averaged result from Equation 1, eliminating differences in the dimensions of the monitored values ​​(e.g., monitored variable i is strain, unit με; monitored variable j is displacement, unit mm), retaining only the relative magnitude of the correlation strength, ensuring that the correlation strength of different types of monitored variables can be compared. Equation 1 reflects the absolute correlation strength of the i-th and j-th monitored variables within the time window w, retaining the dimensions of the original data (e.g., wavelength parameters, unit nm), and the numerical value is directly related to the amplitude of the monitored variables. Equation 2 standardizes the absolute correlation strength, eliminating differences in dimensions (e.g., different wavelength ranges of fiber optic sensors in different measurement areas), retaining only the relative strength of the correlation.

[0044] In this embodiment, the focus is not only on whether the fiber optic sensor readings themselves exceed the limits, but also on whether the cooperative working mode between them is disrupted. Anomalies in the tunnel segment structure (such as cracks or loose bolts) often disrupt this cooperative relationship first, allowing this method to detect potential problems earlier and more sensitively than single-point monitoring. Because absolute correlation strength retains dimensions and sign, it reflects the degree of physical impact of anomalies. Dramatic changes in its value (especially sign reversal) can be directly linked to the severity of physical events (such as stress release or load path changes). Due to the dimensionless nature of relative correlation strength, it is well-suited for system-level, cross-measurement area overall health assessment. A global relative correlation strength matrix can be constructed, and the stability of the entire tunnel segment structure can be judged by observing the overall pattern changes of this matrix. Furthermore, by implementing the calculation of absolute and relative correlation strengths, this invention transforms the raw, massive, low-level time-series data into high-level cooperative relationship features with clear physical and mathematical meaning. This is not only data dimensionality reduction but also lays a solid foundation for subsequent anomaly diagnosis, a key step in achieving high-precision, intelligent monitoring of subway tunnel structures.

[0045] In some embodiments of the present invention, such as Figure 4 As shown, it also includes: S010. Obtain normal feature sequences and abnormal feature sequences, and divide them into training sets and validation sets; the normal feature sequences are time series signals collected when the pipe structure is in a healthy state, and the abnormal feature sequences are time series signals collected when the pipe structure is in an unhealthy state; both the normal feature sequences and the abnormal feature sequences are labeled with the monitoring variables. S020. Based on the normal feature sequences and the abnormal feature sequences in the training set, the temporal convolutional network model is trained. S030. Based on a multivariate Gaussian distribution, calculate the abnormality score corresponding to each of the normal feature sequence and the abnormal feature sequence in the validation set. S040. Traverse different candidate thresholds and calculate the accuracy, recall, and harmonic mean of anomaly detection for each candidate threshold; the harmonic mean is calculated based on the accuracy and recall. S050. Determine the highest harmonic mean as the anomaly score threshold.

[0046] It should be noted that the core parameter of the multivariate Gaussian distribution is the mean vector μ. k The covariance matrix Σ is fitted entirely based on the prediction error of the TCN model's training set. The system operates long-term, collecting massive amounts of historical time-series signals. These data originate from signals acquired and demodulated by the fiber optic sensor array in previous steps. Data collected during healthy tunnel operation periods (e.g., after construction or major repairs, when no defects are confirmed) is extracted and labeled as normal, resulting in normal feature sequences. Data collected during periods of known defects (e.g., after cracks or settlement, confirmed through manual inspection records) is also extracted, or generated by injecting simulated abnormal patterns (e.g., sudden step jumps, slow drifts) into normal data; these are labeled as abnormal, resulting in abnormal feature sequences. All normal and abnormal feature sequences are clearly labeled with their corresponding monitoring variables (fiber optic sensor IDs). A large database of wavelength parameters for corresponding times across multiple measurement areas is established, containing both normal and abnormal feature sequences. For each set of wavelength parameters, the corresponding displacement and strain characteristics are accurately obtained and labeled using specialized measurement equipment and analysis methods. The database covers a large amount of data under different geological conditions, different operating states, and different segment structure types to ensure the diversity and comprehensiveness of the data and provide a solid data foundation for subsequent model training.

[0047] The entire labeled dataset (containing normal and abnormal feature sequences) is randomly divided according to a certain ratio, for example, 70% as the training set and 30% as the validation set. Model hyperparameters are configured, such as kernel size, number of convolutional layers, dilation coefficient, and number of hidden layer neurons. Normal and abnormal feature sequences (i.e., multivariate time series signals) from the training set are used as input. Using the backpropagation algorithm and optimizer (such as Adam), the hyperparameters of the TCN model are continuously adjusted by minimizing the loss function (such as mean squared error) between the predicted and true values. Furthermore, the trained model is used to make predictions on the validation set. After training, a well-trained TCN model that accurately understands the changing patterns of normal data is obtained. In summary, leveraging the excellent temporal modeling capabilities of the Temporal Convolutional Network (TCN) and its ability to capture long-term dependencies and local features in the data, various wavelength parameters and corresponding displacement and strain feature sequences contained in a large labeled database are input into the Temporal Convolutional Network (TCN) for training. During training, the model parameters, such as kernel size, number of convolutional layers, and learning rate, are continuously adjusted to enable the model to accurately learn the mapping relationship between wavelength parameters and displacement and strain characteristic sequences, and to establish a complex relationship model between the deformation of the tunnel segment structure and various influencing factors.

[0048] A multivariate Gaussian distribution model is used to fit the prediction error of the model on the training set. The resulting multivariate Gaussian distribution model is then used to calculate anomaly scores on the validation set to determine the anomaly score threshold. Specifically, the trained TCN model is used to predict validation set data (containing known normal and anomalous samples), and the anomaly score for each validation sample is calculated. The anomaly score calculation process involves first averaging the M error vectors element-wise to obtain the mean vector μ. k , (k=1~n, where n is the number of monitored variables); then, calculate the covariance matrix Σ. ij , where the covariance matrix Σ ij The covariance matrix Σ reflects the correlation between the errors of the i-th and j-th monitoring variables. ij The calculation formula is: .

[0049] Next, verify the fitting effect, that is, verify whether the error vector conforms to the multivariate Gaussian distribution by chi-square test or QQ plot (if the deviation is large, the TCN model parameters need to be optimized and then refitted).

[0050] Next, the Mahalanobis Distance algorithm is used to calculate the anomaly scores corresponding to the normal feature sequences in the validation set, and a candidate threshold range is set (e.g., from 0.5 to 10, with a step size of 0.1). Different candidate thresholds are iterated through, and the anomaly detection accuracy, recall, and F1 score are calculated for each threshold. Among all candidate thresholds, the one that results in the highest F1 score is selected as the final anomaly score threshold (e.g., threshold = 3.2) – ensuring a balance between reducing false negatives (high recall) and reducing false positives (high accuracy), meeting the safety requirement of subway monitoring: err on the side of caution against false alarms.

[0051] In this embodiment, a systematic grid search and F1 score optimization based on validation set data ensure that the final anomaly score threshold has a solid statistical basis. The decision-making process is highly objective, repeatable, and verifiable. By maximizing the F1 score, a harmonic metric, the anomaly score threshold is automatically found, achieving an optimal trade-off between safety and economy. Furthermore, by using a validation set independent of the training set to evaluate performance and determine the threshold, overfitting of the model to the training data is effectively avoided. This ensures that the trained model and the determined threshold also have excellent judgment capabilities for new and unseen data in the future, guaranteeing the long-term reliability and stability of the system.

[0052] In some embodiments of the present invention, such as Figure 5 As shown, step S300 includes: S310. Input the time series signal into the temporal convolutional network model to obtain the predicted value of the monitored variable at the next time step; S320. Calculate the error vector based on the predicted value and the actual measured value of the monitored variable; S330. Based on the multivariate Gaussian distribution and Mahalanobis distance algorithm, the anomaly score is calculated according to the error vector.

[0053] It should be noted that the system acquires historical time series data of length w up to the current time t in real time. This data window [tw, t] contains the readings of all n monitored variables. This data is then formatted into a standardized tensor, for example, of shape (1, n, w+1), representing (batch size, number of variables, time step), respectively. The prepared input tensor is then fed into a pre-trained Temporal Convolutional Network (TCN) model. Internally, the TCN model performs deep analysis of the temporal patterns of the input sequence through its multi-layered causal convolution and dilated convolution structures. Based on the "patterns" learned from massive amounts of normal data, the TCN model infers the value that each monitored variable should have at the next time step (t+1) under "normal conditions." The TCN model outputs an n-dimensional prediction vector. .in, This is the predicted value of the i-th monitored variable at time t+1. At time t+1, the fiber optic sensor demodulator actually collects the readings of all fiber optic sensors, obtaining the actual measurement value vector. The error vector is obtained by subtracting the predicted value from the actual value element by element. , where the error vector The calculation is shown in the following formula: ; .

[0054] Then, the multivariate Gaussian distribution parameters (i.e., mean vector and covariance matrix) calculated based on normal data during the training phase are loaded from system storage. Anomaly scores are then calculated using the Mahalanobis distance algorithm based on the mean vector and covariance matrix. The anomaly score is calculated using the following formula: ; Where, μ k It is the mean vector of the prediction errors of the model training set, Σ ij It is the covariance matrix of the error, μ k and Σ ij Obtained by fitting a multivariate Gaussian distribution. Here, represents the abnormal score, and T is the symbol for calculating the transpose of the matrix.

[0055] In this embodiment, real-time monitoring data is input into a trained temporal convolutional network (TCN). Based on the learned patterns, the network judges the current trend of the tunnel segment structure, analyzing whether the structure is in a stable state, a slowly changing state, or a rapidly deforming state. The TCN model's prediction of "normal" is accurate; therefore, when extremely weak early damage occurs, the resulting tiny error is amplified by the model (because the model expects it to be normal). Furthermore, through comprehensive quantization using the Mahalanobis distance algorithm, these weak abnormal signals distributed across different fiber optic sensors are converged and amplified, enabling reliable detection and achieving true early warning. Moreover, the calculation of the error vector and the Mahalanobis distance algorithm involves simple matrix operations with low computational overhead. This allows the entire scoring process to run in real-time at a high frequency (e.g., once per minute), meeting the stringent real-time monitoring requirements of subway operations.

[0056] In some embodiments of the present invention, step S400 includes: S410. If the correlation change rate between the monitored variables calculated according to the correlation matrix is ​​greater than or equal to the preset change rate, and the change rate of the abnormal score is greater than or equal to the preset rate, the abnormality type is determined to be a sudden phase abnormality. S420. If the rate of change of the correlation between the monitored variables calculated according to the correlation matrix is ​​less than the preset rate of change, and the rate of change of the abnormal score is less than the preset rate, the abnormality type is determined to be a gradual phase abnormality.

[0057] It should be noted that if the phase of the entire monitored variable suddenly rises or falls due to an unexpected event, or when the entire phase changes due to a concept drift or change point, an anomaly can be detected by calculating the anomaly score and predicting the error vector. If the anomaly score of a point is greater than a threshold, an anomaly can be detected. Each fiber optic sensor corresponds to one monitored variable; the fiber optic sensor demodulator collects wavelength parameters from multiple measurement areas at corresponding times in real time and demodulates them into a continuous time-series signal x. i (For example, recording strain values ​​every 10 minutes to form a continuous time-strain sequence). Only by acquiring this continuous time series data can we observe phase rise / fall (e.g., strain rising from 20 με to 50 με) or phase trend changes (e.g., strain changing from stable fluctuations to a continuous rise). Phase changes are identified through correlation matrix + outlier score (data preprocessing and feature extraction). That is, phase changes can be accurately captured through correlation matrix analysis and outlier score calculation, which is specifically divided into two steps: Step 1: Using the correlation matrix mt Capturing phase anomalies in variable co-correlation involves calculating the correlation matrix at time t from the correlation between two sequences. mt ∈ Rn × n —Under normal circumstances, there is a stable cooperative relationship among the monitoring variables of the tunnel segment structure (such as the joint strain and the body displacement rising and falling synchronously, and the correlation is high). mijt (Stable at 0.8~0.9); If the phase changes due to unexpected events, turning points, or concept drift, it will directly disrupt this cooperative relationship, manifesting as abrupt or gradual changes in the elements of the correlation matrix. The specific judgment logic is as follows: The anomaly type is an unexpected event / change point (sudden phase change or sudden phase anomaly): If an unexpected event such as segment cracking or bolt loosening occurs, or a change point appears (such as a sudden increase in strain), it will cause a sudden mismatch in the originally coordinated monitoring variables—for example, the joint strain (i=1) suddenly increases from 20με to 80με due to the crack (sudden phase increase), but the body displacement (j=5) remains stable at 0.2mm due to the interruption of stress transmission (unchanged phase). At this time, the correlation m15t between the two will suddenly drop from 0.8 to below 0.1. By comparing the real-time correlation matrix mt with the mean of the historical normal correlation matrix, if the difference exceeds the preset threshold (such as difference > 0.5), it can be determined that there is a sudden phase change.

[0058] The anomaly type is conceptual drift (gradual phase change or gradual phase anomaly): If conceptual drift is caused by slow geological subsidence or material aging (e.g., strain slowly increases by 0.5 με per month, phase change is gradual), it will lead to a slow imbalance in the synergistic relationship between monitored variables—for example, the correlation m15t between joint strain (i=1) and bulk displacement (j=5) decreases from 0.8 to 0.05 per month, and drops to 0.65 after 3 months. By continuously tracking the temporal trend of the correlation matrix, if it is found that the correlation of a certain pair of monitored variables changes continuously at a rate exceeding the normal fluctuation range (e.g., monthly decrease > 0.02), it can be determined that there is a gradual phase change (conceptual drift).

[0059] Step 2: Quantify the degree of anomalousness of phase changes using anomaly scores: If the phase of the entire monitored variable suddenly rises or falls due to an unexpected event, or when the entire phase changes due to concept drift or a change point, an anomaly can be detected by calculating the anomaly score and predicting the error vector. If the anomaly score of a point is greater than a threshold, an anomaly can be detected. The core quantitative method for judging phase changes relies on TCN model prediction + multivariate Gaussian distribution fitting, and the steps are as follows (combined with the real-time recognition stage of model training): During the model training phase, a baseline for the error distribution of normal phase is established. First, a database of wavelength parameters containing normal / abnormal feature sequences is constructed, labeled with corresponding displacement and strain characteristics. The TCN model is used to learn the phase patterns of normal data (e.g., the fluctuation range and rate of change of the strain sequence under normal conditions), and the prediction error of the TCN model on the training set is fitted using a multivariate Gaussian distribution to obtain the mean vector μ of the error. k The covariance matrix Σ—these two parameters define the prediction error range corresponding to the normal phase (i.e., the error within the 99% confidence interval is considered normal). In the real-time identification stage, real-time monitoring data is input into the trained TCN model, and phase changes are determined through the following logic: Calculate the prediction error vector—the TCN model predicts the normal phase data that should exist at the current moment (e.g., the predicted strain should be 22 με), compares it with the real-time monitored phase data (e.g., the actual strain is 40 με), and obtains the difference between the two, i.e., the prediction error vector. The outlier score is calculated using a Mahalanobis distance algorithm based on a multivariate Gaussian distribution. If the outlier score exceeds a preset threshold (determined through a validation set to ensure that outlier scores for normal data remain within the threshold), it indicates that the real-time phase deviates significantly from the normal phase. A sudden increase in the outlier score (e.g., from 1.2 to 5.8) corresponds to a sudden phase change caused by an unexpected event / turning point. A gradual accumulation of outlier scores (e.g., from 1.0 to 3.5 over a month) corresponds to a gradual phase change caused by concept drift.

[0060] In this embodiment, by requiring two different dimensions of indicators (coordination relationship and overall pattern) to simultaneously meet specific change patterns, the system can effectively eliminate many single, sporadic interference signals. For example, a brief failure of a single fiber optic sensor may cause an anomaly score to rise instantaneously, but it usually does not cause a sudden change in the correlation matrix, thus avoiding misjudgment as sudden structural damage. Furthermore, different types of anomalies (sudden / gradual) often correspond to different physical damage mechanisms. The system's automatic identification of anomaly types provides valuable, categorized field data for backend analysts to study structural degradation patterns and optimize segment design, forming a closed loop from monitoring to understanding.

[0061] like Figure 6 As shown, this invention is based on signals acquired by a fiber optic sensor array. By introducing the TCN (Temporal Convolutional Network) deep learning algorithm, a temporal convolutional network framework is used as the prediction model to capture the temporal relationships and long-term dependencies in the strain feature sequence through convolution operations. By stacking multiple convolutional layers, higher-level strain features are extracted layer by layer, and a multivariate Gaussian distribution is used to identify anomalies in the time series (i.e., the strain feature sequence), achieving accurate strain feature sequence prediction. The scheme mainly includes a data acquisition layer, a preprocessing and feature extraction layer, a temporal pattern learning and classification layer, and a result analysis and alarm layer. This invention enables the monitoring of subway tunnel segment structures using a fiber optic sensor array vibration cable. The vibration measurement zone formed between every two adjacent grating vibration fiber optic sensors can quickly detect structural anomalies. Compared with traditional point-type fiber optic sensors, this greatly improves the monitoring accuracy and resolution, and can promptly detect minute deformations in the segment structure. Furthermore, by collecting segment structure deformation data and related influencing factor data over a long period and using a large AI model for deep analysis, hidden patterns and trends in the data can be uncovered, accurately judging the changing trends of the segment structure and providing strong support for disease prediction. Furthermore, by combining structural mechanics principles with historical data, the AI-powered large-scale model can accurately predict the timing of segment structure defects, providing early warnings to subway tunnel maintenance and management departments. This allows them sufficient time to develop maintenance plans and take appropriate measures, reducing the risks and losses caused by defects. Moreover, by integrating the AI-powered large-scale model results with domain expert knowledge, the constructed segment structure monitoring expert model exhibits higher reliability and practicality. Expert experience and knowledge can compensate for potential model shortcomings, further improving the accuracy of defect prediction and the scientific rigor of decision support.

[0062] To better implement the subway tunnel segment structure monitoring method in this invention, based on the subway tunnel segment structure monitoring method, correspondingly, as follows: Figure 7 As shown, this embodiment of the invention also provides a subway tunnel segment structure monitoring system. The subway tunnel segment structure monitoring system 700 includes: The acquisition module 701 is used to acquire time series signals including at least two monitoring variables; the time series signals are collected by an array of fiber optic sensors laid on the inner wall of the subway tunnel. Construction module 702 is used to construct a correlation matrix between different monitored variables at the current time based on the time series signal; the correlation matrix is ​​used to characterize the cooperative relationship between different monitored variables; Prediction module 703 is used to obtain the corresponding anomaly score based on the time series signal according to a pre-trained temporal convolutional network model. The generation module 704 is used to analyze whether the segment structure has any abnormalities based on the correlation matrix and the anomaly score.

[0063] The subway tunnel segment structure monitoring system 700 provided in the above embodiments can realize the technical solutions described in the above embodiments of the subway tunnel segment structure monitoring method. The specific implementation principles of each module or unit can be found in the corresponding content in the above embodiments of the subway tunnel segment structure monitoring method, which will not be repeated here.

[0064] like Figure 8 As shown, the present invention also provides an electronic device 800. The electronic device 800 includes a processor 801, a memory 802, and a display 803. Figure 8 Only some components of the electronic device 800 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0065] In some embodiments, processor 801 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 802 or process data, such as the subway tunnel segment structure monitoring method of the present invention.

[0066] In some embodiments, processor 801 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 801 may be local or remote. In some embodiments, processor 801 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0067] In some embodiments, memory 802 may be an internal storage unit of electronic device 800, such as a hard disk or memory of electronic device 800. In other embodiments, memory 802 may also be an external storage device of electronic device 800, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 800.

[0068] Furthermore, the memory 802 may include both internal storage units of the electronic device 800 and external storage devices. The memory 802 is used to store application software and various types of data installed on the electronic device 800.

[0069] In some embodiments, display 803 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 803 is used to display information from electronic device 800 and to display a visual user interface. Components 801-803 of electronic device 800 communicate with each other via a system bus.

[0070] In one embodiment, when the processor 801 executes the subway tunnel segment structure monitoring program in the memory 802, the following steps can be implemented: Acquire time-series signals including at least two monitored variables; the time-series signals are collected by an array of fiber optic sensors laid on the inner wall of the subway tunnel; Based on the time series signal, a correlation matrix is ​​constructed between the different monitored variables at the current time; the correlation matrix is ​​used to characterize the cooperative relationship between the different monitored variables. Based on a pre-trained temporal convolutional network model, the corresponding anomaly score is obtained according to the time series signal; Based on the correlation matrix and the anomaly score, an analysis is conducted to determine whether any anomalies have occurred in the segment structure.

[0071] It should be understood that when the processor 801 executes the subway tunnel segment structure monitoring program in the memory 802, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0072] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 800 mentioned. Electronic device 800 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 800 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0073] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the subway tunnel segment structure monitoring method provided in the above-described method embodiments.

[0074] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0075] The above provides a detailed description of the method, system, equipment, and medium for monitoring the structure of subway tunnel segments provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for monitoring the structure of subway tunnel segments, characterized in that, include: Acquire time-series signals that include at least two monitored variables; The time-series signal was acquired by an array of fiber optic sensors laid on the inner wall of the subway tunnel. Based on the time series signal, a correlation matrix is ​​constructed between the different monitored variables at the current time; the correlation matrix is ​​used to characterize the cooperative relationship between the different monitored variables. Based on a pre-trained temporal convolutional network model, the corresponding anomaly score is obtained according to the time series signal; Based on the correlation matrix and the anomaly score, an analysis is conducted to determine whether any anomalies have occurred in the segment structure.

2. The method for monitoring the structure of subway tunnel segments according to claim 1, characterized in that, The acquisition of time-series signals including at least two monitored variables includes: Acquire the physical signal collected by each fiber optic sensor in the fiber optic sensor array; the physical signal includes the wavelength parameter of the fiber optic sensor; the wavelength parameter corresponds to the monitored variable; The wavelength parameters of the multiple physical signals are demodulated into the time-series signal according to the time order and the number of the fiber optic sensor.

3. The method for monitoring the structure of subway tunnel segments according to claim 1, characterized in that, The construction of a correlation matrix between different monitored variables at the current time based on the time series signal includes: Extract all data of each monitored variable within the same target time window to obtain the corresponding time segment; the target time window is represented as [tw, t], where t is the current time and w is the preset window length; Calculate the average correlation strength of any two different monitored variables corresponding to the time series segments; The average correlation strength corresponding to each of the two different monitored variables is filled into an n×n matrix to obtain the correlation matrix corresponding to the current time, where n is a positive integer greater than 2.

4. The method for monitoring the structure of subway tunnel segments according to claim 3, characterized in that, The calculation of the average correlation strength of any two different monitored variables corresponding to the time series includes: The absolute correlation strength of any two different monitored variables corresponding to the time series segments is calculated: ； or, The relative correlation strength of any two different monitored variables corresponding to the time series segments is calculated: ； in, The absolute correlation strength between i and j Let represent the relative correlation strength between i and j, where i and j are two distinct monitored variables, and n is the total number of monitored variables. Let w be an integer from 0 to w, and ||·|| be the L2 norm function.

5. The method for monitoring the structure of subway tunnel segments according to claim 1, characterized in that, Also includes: Normal feature sequences and abnormal feature sequences are obtained and divided into training sets and validation sets; the normal feature sequences are time series signals collected when the pipe structure is in a healthy state, and the abnormal feature sequences are time series signals collected when the pipe structure is in an unhealthy state. Both the normal feature sequences and the abnormal feature sequences are labeled with the monitoring variables. The temporal convolutional network model is trained based on the normal feature sequences and the abnormal feature sequences in the training set. Based on a multivariate Gaussian distribution, calculate the abnormality score corresponding to each of the normal feature sequence and the abnormal feature sequence in the validation set. Iterate through different candidate thresholds and calculate the accuracy, recall, and harmonic mean of anomaly detection for each candidate threshold; the harmonic mean is calculated based on the accuracy and recall. The highest harmonic mean is determined as the anomaly score threshold.

6. The method for monitoring the structure of subway tunnel segments according to claim 5, characterized in that, The temporal convolutional network module obtains the corresponding anomaly score based on the time series signal, including: The time series signal is input into the temporal convolutional network model to obtain the predicted value of the monitored variable at the next time step. An error vector is calculated based on the predicted and actual measured values ​​of the monitored variables; The anomaly score is calculated based on the error vector using a multivariate Gaussian distribution and Mahalanobis distance algorithm.

7. The method for monitoring the structure of subway tunnel segments according to claim 1, characterized in that, The step of analyzing whether the segment structure exhibits anomalies based on the correlation matrix and the anomaly score includes: If the correlation change rate between the monitored variables calculated based on the correlation matrix is ​​greater than or equal to a preset change rate, and the change rate of the abnormal score is greater than or equal to a preset rate, the abnormality type is determined to be a sudden phase abnormality. If the rate of change of the correlation between the monitored variables calculated based on the correlation matrix is ​​less than the preset rate of change, and the rate of change of the abnormal score is less than the preset rate, the abnormality type is determined to be a gradual phase abnormality.

8. A monitoring system for subway tunnel segment structure, characterized in that, include: The acquisition module is used to acquire time series signals including at least two monitoring variables; the time series signals are collected by an array of fiber optic sensors laid on the inner wall of the subway tunnel. The construction module is used to construct a correlation matrix between different monitored variables at the current time based on the time series signal; the correlation matrix is ​​used to characterize the cooperative relationship between different monitored variables. The prediction module is used to obtain the corresponding anomaly score based on the time series signal, using a pre-trained temporal convolutional network model. The generation module is used to analyze whether the segment structure has any abnormalities based on the correlation matrix and the anomaly score.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the subway tunnel segment structure monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the subway tunnel segment structure monitoring method according to any one of claims 1 to 7.

Images