Earth surface real-time monitoring system based on shield construction tunnel

Abstract

The invention relates to the technical field of ground surface settlement monitoring, and discloses a ground surface real-time monitoring system based on a shield construction tunnel. A data acquisition unit of the system obtains a displacement measurement value from an earth surface sensor and generates time sequence data; the sequence segmentation unit analyzes mutation points in the sequence and segments the sequence into a plurality of data segments according to the mutation points; an exception marking unit performs statistical test on each data segment to identify an exception data segment set; the trend evaluation unit performs trend line fitting on the abnormal set and evaluates the fluctuation degree of fitting deviation; the matrix conversion unit organizes the time sequence data into a specific matrix structure and performs matrix decomposition; a feature derivation unit derives feature parameters from the decomposition result; and the monitoring control unit adaptively adjusts a monitoring strategy according to the characteristic parameters and generates a settlement report. According to the system, accurate and self-adaptive monitoring and early warning of ground surface settlement are realized through multi-stage data processing and intelligent analysis.

Description

Technical Field

[0001] This invention relates to the field of surface settlement monitoring technology, specifically a real-time surface monitoring system based on shield tunnel construction. Background Technology

[0002] In subway construction, tunnel boring machine (TBM) construction has become the primary method due to its unique advantages. TBM construction is characterized by its speed and efficiency; the TBM can operate continuously underground, enabling rapid tunnel excavation and significantly shortening the construction cycle. Compared to traditional open-cut methods, TBM construction has less impact on surface traffic and the surrounding environment, eliminating the need for large-scale surface excavation and reducing disruption to normal urban life. The high degree of mechanization in TBM construction effectively ensures the quality and precision of tunnel construction, reduces construction errors, and improves tunnel stability and safety.

[0003] Tunnel boring machine (TBM) construction also presents several challenges. During construction, the cutterhead of the TBM continuously cuts into the soil, disturbing the surrounding strata and altering their original stress state. This disturbance can lead to changes in the structure between soil particles, causing the soil to loosen and deform, potentially resulting in surface heave or subsidence. If surface subsidence or heave exceeds a certain limit, it can pose a serious threat to surface buildings and underground pipelines. Surface buildings may experience wall cracking, foundation subsidence, and structural tilting, affecting their normal use and safety. Underground pipelines may rupture or deform, causing interruptions to water, electricity, and gas supply systems, greatly disrupting the normal operation of the city.

[0004] To promptly monitor surface changes during tunnel boring machine (TBM) construction and ensure project safety and environmental stability, real-time surface monitoring is essential. However, existing TBM monitoring technologies have several shortcomings. Current monitoring methods often fail to acquire surface change information in real time; data acquisition involves time delays, resulting in monitoring results that do not promptly reflect the actual surface conditions. Traditional manual monitoring methods are not only infrequent but also susceptible to human error, leading to poor data accuracy and reliability. While some automated monitoring equipment can increase monitoring frequency, delays may still occur during data transmission and processing, failing to meet the requirements of real-time monitoring.

[0005] Due to the inability to obtain accurate information on surface changes in a timely manner, construction personnel struggle to adjust tunnel boring machine (TBM) parameters promptly. In TBM construction, soil chamber pressure and grouting volume are two critical parameters that play a vital role in controlling surface settlement. Improperly set soil chamber pressure—too high or too low—can lead to soil instability, causing surface settlement or heave. Insufficient or uneven grouting volume results in loose soil filling, also causing surface settlement. However, existing monitoring systems cannot provide construction personnel with accurate surface settlement data in a timely manner, making it difficult to adjust soil chamber pressure and grouting volume based on actual conditions. This hinders effective control of surface settlement and increases project risks. Summary of the Invention

[0006] The purpose of this invention is to provide a real-time surface monitoring system for shield tunnel construction to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a real-time surface monitoring system based on shield tunnel construction, the system comprising: The system includes a data acquisition unit, a sequence segmentation unit, an anomaly marking unit, a trend assessment unit, a matrix transformation unit, a feature derivation unit, and a monitoring and control unit. The data acquisition unit is configured to acquire displacement measurements from surface sensors and generate time-series data; the sequence segmentation unit is configured to analyze abrupt changes in the time-series data and segment the sequence into multiple data segments based on these abrupt changes; the anomaly marking unit is configured to perform statistical tests on each data segment to identify a set of anomalous data segments; the trend evaluation unit is configured to fit a trend line to the set of anomalous data segments and evaluate the degree of fluctuation in the fitting deviation; the matrix transformation unit is configured to organize the time-series data into a specific matrix structure and perform matrix decomposition; the feature derivation unit is configured to derive feature parameters from the matrix decomposition results; and the monitoring and control unit is configured to adjust the monitoring strategy based on the feature parameters and generate a settlement report.

[0008] Preferably, the sequence segmentation unit uses a sliding window technique to scan the time series data and calculates the variance of the displacement values ​​within each window; when the rate of change of variance between consecutive windows exceeds a preset threshold, the position is recorded as a mutation point; the time series data is divided into independent data segments from each mutation point; the displacement range and standard deviation of each data segment are calculated, and the product of the range and standard deviation is used as the fluctuation intensity of the data segment; a grouping algorithm is used to cluster the fluctuation intensity of all data segments, and the data segment corresponding to the group with the largest fluctuation intensity is selected as a candidate abnormal data segment.

[0009] Preferably, the anomaly marking unit calculates the similarity measure between each candidate anomaly data segment and its preceding and following adjacent data segments. The similarity measure is based on the distribution pattern of displacement values ​​in the data segment. A similarity threshold is set, and data segments with similarity below the threshold are initially marked as suspicious anomaly segments. The length of the suspicious anomaly segments is verified, and only data segments with a length greater than the minimum valid duration are retained as the final set of anomaly data segments.

[0010] Preferably, the trend assessment unit performs polynomial regression fitting on each final abnormal data segment to obtain a trend curve; calculates the residual sequence between the trend curve and the actual displacement value; extracts the peak and valley values ​​of the residual sequence, and calculates the average distance between the peak and valley values ​​as the local fluctuation amplitude; integrates the local fluctuation amplitudes of all final abnormal data segments, and obtains the overall trend fluctuation characteristics by weighted averaging.

[0011] Preferably, the matrix transformation unit arranges the time series data into a Toplitz matrix, where each row of the Toplitz matrix represents the displacement sequence of a lag time window; performs eigenvalue decomposition on the Toplitz matrix to obtain an eigenvalue set; arranges the eigenvalues ​​in descending order and calculates the ratio sequence between consecutive eigenvalues; identifies jump points in the ratio sequence and determines the eigenvalues ​​before the jump points as the main eigenvalue group.

[0012] Preferably, the feature derivation unit separates the primary feature value group and the secondary feature value group from the feature value set; calculates the ratio of the cumulative sum of the primary feature value group to the sum of the entire feature value set as the energy concentration; calculates the relative difference between the variance of the primary feature value group and the variance of the secondary feature value group as the feature dispersion index; and derives feature parameters by combining the energy concentration and the feature dispersion index.

[0013] Preferably, the monitoring and control unit multiplies the energy concentration and characteristic dispersion index to obtain a signal reliability score; dynamically adjusts the data sampling frequency and report generation interval according to the signal reliability score; and triggers a high-frequency monitoring mode and sends a real-time alarm when the signal reliability score is lower than a safety threshold.

[0014] Preferably, the data acquisition unit includes multiple surface displacement sensors, which are arranged in a grid according to geographic coordinates; the data acquisition unit collects sensor data at regular intervals and applies digital filters to smooth data fluctuations; a linear interpolation method is used to compensate for missing data points and output continuous time series data.

[0015] Preferably, the system further includes a data calibration unit for periodically calibrating sensor data against a reference point; the data calibration unit calculates the systematic error between the sensor data and the reference point, and applies an error compensation algorithm to adjust the displacement measurement value.

[0016] Preferably, the monitoring and control unit generates a visual report containing the location of abnormal data segments and trend assessment results; when the characteristic dispersion index rises abnormally, it automatically activates the redundant sensor network for cross-validation and outputs detailed analysis results to the monitoring platform.

[0017] Compared with the prior art, the beneficial effects of the present invention are: The data acquisition unit of this invention obtains displacement measurements from surface sensors and generates time-series data. This data covers detailed information on the changes in surface displacement over time during shield tunneling, providing a comprehensive and accurate foundation for subsequent analyses. The sequence segmentation unit divides the time-series data into multiple segments based on abrupt change points, enabling more detailed and in-depth data analysis. Different data segments may correspond to different stages of shield tunneling or different geological conditions. By studying these data segments separately, the patterns and characteristics of surface changes can be grasped more accurately. The anomaly marking unit performs statistical tests on each data segment to identify a set of anomalous data segments, which helps to promptly identify time periods and areas where surface displacement anomalies occur. The trend assessment unit performs trend line fitting and fluctuation assessment on the set of anomalous data segments, further analyzing the changing trends of the anomalous data segments and providing more valuable references for subsequent decision-making. The matrix transformation unit organizes the time-series data into a specific matrix structure and performs matrix decomposition, processing and analyzing the data from another perspective to uncover hidden features and information within the data. The feature-derived unit derives feature parameters from the matrix decomposition results. These feature parameters can more comprehensively reflect the changing characteristics of surface displacement, providing richer decision-making basis for the monitoring and control unit.

[0018] The anomaly marking unit uses statistical testing methods to analyze data segments, accurately identifying sets of anomalous data segments. This statistically based method boasts high accuracy and reliability, effectively avoiding misjudgments and omissions. Timely identification of anomalous data segments allows construction personnel to quickly identify potentially risky areas and time periods, taking appropriate measures such as adjusting tunnel boring machine (TBM) parameters and strengthening the protection of surrounding buildings and underground pipelines, thereby reducing project risks and ensuring construction safety. The trend assessment unit performs trend line fitting and fluctuation assessment on the set of anomalous data segments, visually displaying their changing trends and fluctuations. Construction personnel can use these assessment results to predict the development trend of surface displacement and prepare in advance. If the trend assessment shows a continuous increase in surface displacement, construction personnel can promptly adjust TBM parameters, such as reducing the advance speed and increasing the grouting volume, to control the development of surface displacement and avoid serious safety accidents.

[0019] The feature-derived unit derives feature parameters from the matrix decomposition results. These parameters comprehensively reflect various characteristics of surface displacement, such as trends, fluctuations, and periodicity. The monitoring and control unit adjusts its monitoring strategy based on these parameters, achieving a scientific and rational approach. When the feature parameters indicate relatively stable surface displacement, the monitoring and control unit can appropriately reduce the monitoring frequency to lower costs. Conversely, when abnormal changes occur, the unit can promptly increase the monitoring frequency to strengthen monitoring and ensure timely detection and intervention. The unit also generates settlement reports, detailing relevant information such as settlement amount, rate, and area, providing construction and management personnel with comprehensive and accurate information on surface settlement, facilitating decision-making and management. These reports can also be stored as engineering data, providing a reference for subsequent maintenance and renovation. Attached Figure Description

[0020] Figure 1 This is a schematic diagram illustrating the working principle of the real-time surface monitoring system based on shield tunneling as described in this invention. Figure 2 Flowchart illustrating the working principle of a sequence segmentation unit; Figure 3 A flowchart illustrating the working principle of the trend assessment unit; Figure 4 A graph showing the changing trends of surface monitoring characteristic parameters during shield tunneling construction; Figure 5 A trend chart showing the dynamic adjustment of the reliability of surface monitoring signals and sampling frequency during shield tunneling construction. Detailed Implementation

[0021] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1This invention provides a real-time surface monitoring system for shield tunneling. The system includes: a data acquisition unit, a sequence segmentation unit, an anomaly marking unit, a trend assessment unit, a matrix transformation unit, a feature derivation unit, and a monitoring and control unit. The data acquisition unit acquires displacement measurements from a sensor network deployed on the surface and generates continuous time-series data, which is filtered and interpolated to ensure integrity. The sequence segmentation unit receives the time-series data and divides the sequence into multiple independent data segments by analyzing abrupt changes, thus providing a structured basis for subsequent analysis. The anomaly marking unit performs statistical tests on each data segment to identify a set of anomalous data segments, ensuring the accuracy of the anomaly marking through similarity measurement and length verification. The trend assessment unit fits trend lines to the set of anomalous data segments and evaluates the degree of fluctuation in the fitting deviation, thereby extracting local and overall trend features. The matrix transformation unit organizes the time-series data into a matrix structure, such as the Topplitz matrix, and performs matrix decomposition to reveal the inherent patterns in the data. The feature derivation unit derives key feature parameters from the matrix decomposition results, including energy concentration and feature dispersion indices. The monitoring and control unit dynamically adjusts the monitoring strategy based on characteristic parameters, such as adjusting the sampling frequency and report generation interval, and generates settlement reports to achieve intelligent monitoring and control. All units are connected in series via data streams to form a closed-loop processing flow, ensuring the system responds to surface changes in real time.

[0023] Example 1: See Figure 2 In practice, the width of the sliding window is preset according to the actual needs of engineering monitoring. For example, it can be set to cover the number of sampling points from tens of minutes to several hours. The window slides forward along the time axis with a fixed step size, and the variance of all displacement measurements within the new window is calculated after each slide. The parameter settings of the sliding window need to be comprehensively determined in conjunction with the tunnel boring machine construction progress, stratum stability, and monitoring accuracy requirements. The selection of the window width is based on the ability to capture the effective change cycle of stratum displacement. If the construction area is a soft soil stratum, the stratum deformation rate is fast and fluctuations are frequent, the window width can be preset to cover the number of sampling points for 30-60 minutes to ensure timely capture of short-term changes. If it is a hard rock stratum, the deformation is gentle and the cycle is long, the window width can be extended to the number of sampling points for 2-4 hours to avoid oversensitivity leading to misjudgment. The window step size needs to be set in a balance between detection efficiency and computational cost. The step size is typically 1 / 4 to 1 / 2 of the window width. When the window width corresponds to 30 minutes of sampling points, the step size can be set to the number of sampling points over 7-15 minutes, ensuring the continuity of data between adjacent windows while avoiding excessive redundant calculations. When the window width is large, the step size can be increased proportionally to ensure the system can efficiently scan the entire time series data without missing any key abrupt changes. Before selecting parameters, the monitoring data characteristics of similar projects should be referenced. In the initial stages of construction, adjustments can be made through small-scale trial monitoring to ultimately determine fixed parameters suitable for the current engineering geology and construction conditions.

[0024] The variance change rate is calculated by comparing the variance of the current sliding window with that of the previous sliding window. Specifically, it is the ratio of the absolute difference between the variances of two consecutive sliding windows to the average variance of the two sliding windows. When this ratio exceeds a preset threshold, the system determines that there is a sudden change point at that location and records its timestamp. The preset threshold is determined based on the normal fluctuation range of ground displacement during shield tunneling, combined with historical monitoring data and engineering safety standards. First, complete monitoring data of similar shield tunneling projects in the past are collected, and time series data of construction stages without abnormalities are selected. The variance change rate of the consecutive sliding windows in this stage is calculated, and its maximum value and common fluctuation range are statistically analyzed. The upper limit of the normal fluctuation range is used as the initial threshold reference, and then adjusted according to the safety level of the current project. If the construction area is near important buildings or underground pipelines, and the safety requirements are high, it is necessary to strictly capture minor sudden changes, and the initial threshold can be lowered by 10%-20%. If it is a normal construction area, the initial threshold can be maintained or appropriately raised to avoid oversensitivity. Furthermore, the preset process needs to consider different stages of shield tunneling. When tunneling through complex strata, the strata disturbance is large and the normal variance variation rate is high, so the threshold can be appropriately widened. During the stable tunneling stage, the threshold can be tightened. After preset, it needs to be verified through trial monitoring. If it is found that the threshold is too high, resulting in missed abrupt changes, or too low, resulting in too many false alarms, it should be fine-tuned in a timely manner to a reasonable value that is suitable for the current construction conditions.

[0025] At each identified abrupt change, the time series data is segmented into a series of independent data segments that are continuous in time but may differ in statistical characteristics. After segmentation, the sequence segmentation unit calculates the displacement range and standard deviation for each independent data segment. The displacement range is the arithmetic difference between the maximum and minimum displacement values ​​in the data segment, and the specific formula is as follows:

[0026] Where R represents the displacement range. This represents the maximum value among the displacement measurements within the data segment. This represents the minimum value among the displacement measurements within the data segment.

[0027] The standard deviation of displacement characterizes the degree of dispersion of displacement values ​​within a data segment relative to its average value. The specific formula for calculation is as follows:

[0028] in, represents the standard deviation of displacement, and n represents the number of displacement measurements within the data segment. This represents the i-th displacement measurement value within the data segment. The arithmetic mean of all displacement measurements within the data segment ( ).

[0029] Multiplying the calculated displacement range by the displacement standard deviation, the product is defined as the fluctuation intensity of the data segment. Fluctuation intensity is a dimensionless quantitative indicator used to comprehensively reflect the severity and stability of displacement changes within the data segment. The specific formula for calculating fluctuation intensity is as follows:

[0030] Where I represents the wave intensity and R represents the displacement range. This represents the standard deviation of displacement. Fluctuation intensity is a dimensionless quantitative indicator used to comprehensively reflect the severity and stability of displacement changes within a data segment.

[0031] The sequence segmentation unit uses a grouping algorithm to perform cluster analysis on the fluctuation intensity values ​​of all data segments. In some embodiments, the grouping algorithm can adopt the K-means clustering method, which aims to naturally group the fluctuation intensity values ​​according to their numerical values. The group with the largest fluctuation intensity is considered to contain the data segments with the greatest potential for anomalousness, and all data segments corresponding to this group are output as candidate anomalous data segments.

[0032] The anomaly labeling unit receives candidate anomaly data segments from the sequence segmentation unit and calculates a similarity measure for each candidate anomaly data segment. In practice, the similarity measure focuses on comparing the consistency of the displacement value distribution pattern between the candidate anomaly data segment and its direct predecessor and successor data segments. The comparison of distribution patterns can be achieved by calculating the distance between the empirical distribution functions of displacement values ​​between data segments. For example, the Kolmogorov-Smirnov test statistic can be used as the basis for the similarity measure, and its specific formula is as follows:

[0033] in, Let x represent the Kolmogorov-Smirnov statistic for two data segments (sample sizes n and m), where x represents the displacement variable and supremum (maximum value).

[0034] Alternatively, it can be done by comparing the higher-order statistical moments (skewness and kurtosis) of the data segments. The specific formula for skewness is:

[0035] The specific formula for kurtosis is:

[0036] in, Represents the skewness of the data segment. The kurtosis of the data segment. This represents the i-th displacement measurement value in the data segment. The value represents the arithmetic mean of the displacement measurements in the data segment, where n represents the number of displacement measurements within the data segment.

[0037] The system presets a similarity threshold. When the similarity metric between the calculated candidate abnormal data segment and its adjacent data segments is lower than this threshold, the candidate abnormal data segment is initially marked as a suspicious abnormal segment, which suggests that its displacement distribution pattern is significantly different from the surrounding normal evolution pattern.

[0038] The preset similarity threshold is determined based on statistical analysis of historical monitoring data and comprehensive engineering experience. Monitoring data from normal shield tunneling operations in the same region or under similar geological conditions are collected. Continuous data segments without anomalies are extracted, and the similarity metric between these normal data segments is calculated, along with their distribution range and concentration intervals. The minimum similarity metric of the normal data segments is used as the base threshold reference, which is then adjusted based on the sensitivity requirements for anomaly identification in the project. If strict risk control and reduced missed detections are required, the base threshold can be appropriately lowered; if frequent false alarms need to be avoided and monitoring stability ensured, the base threshold can be appropriately increased. Simultaneously, the preset process must consider differences in data distribution patterns. For areas with stable geological deformation patterns, the threshold can be set more concentrated. For areas with complex geological conditions and large fluctuations in normal data, the threshold needs to have a certain margin of error. The preset threshold is not fixed and can be continuously calibrated through feedback from subsequent monitoring data to ensure accurate differentiation between normal fluctuations and anomaly patterns.

[0039] After initial labeling, the anomaly labeling unit performs length verification on each suspected anomaly segment. The purpose of length verification is to eliminate short-duration anomaly signals that may be caused by transient interference, thus avoiding false alarms. Length verification is achieved by comparing the duration covered by the suspected anomaly segment with a preset minimum effective duration parameter. The minimum effective duration is typically determined based on the sensor sampling interval and the shortest duration of anomaly events considered meaningful in engineering practice. Only suspected anomaly segments with a duration greater than the minimum effective duration are retained and confirmed as final anomaly data segments, while segments with insufficient length are discarded. All verified final anomaly data segments constitute the final anomaly data segment set and are sent to the trend assessment unit for subsequent analysis. By combining statistical similarity measures and hard constraints on physical duration, the anomaly labeling unit can identify truly valuable anomaly data segments with high reliability.

[0040] In the specific implementation of the sliding window technique, the selection of window width and sliding step size needs to balance detection sensitivity and computational overhead. In some embodiments, the window width may be set to include 128 sampling points, and the sliding step size may be set to 32 sampling points to maintain acceptable computational efficiency while ensuring the accuracy of abrupt change detection. The preset threshold value of the variance change rate needs to be determined by analyzing historical or simulated data. The principle for setting it is to effectively capture substantial step changes in formation displacement while suppressing minor fluctuations caused by measurement noise. A threshold set too low may lead to too many false abrupt changes, while a threshold set too high may miss real gradual anomalies. When calculating the fluctuation intensity of a data segment, the product of the displacement range and the displacement standard deviation serves as a composite index, capable of simultaneously capturing the absolute range and internal variability of displacement changes. Compared to using either index alone, fluctuation intensity can more robustly characterize the anomaly potential of a data segment. In the process of clustering fluctuation intensity using a grouping algorithm, determining the number of clusters is an important parameter. Optionally, the elbow rule or the silhouette coefficient method can be used to automatically determine the optimal number of clusters to ensure that fluctuation intensity values ​​can be reasonably grouped. The group with the strongest fluctuation intensity was selected based on the fundamental assumption that anomalous events during surface subsidence usually trigger the most violent and unstable fluctuation patterns in the displacement sequence.

[0041] In the similarity measurement stage of the anomaly marker unit, comparison is made with directly adjacent preceding and following data segments. This is based on the engineering fact that geological displacement changes typically exhibit continuity and gradual change over time. A genuine anomaly often results in a identifiable shift in the data patterns of its immediate vicinity. The preset similarity threshold needs to be calibrated based on historical data from the specific monitoring environment to effectively distinguish between normal background fluctuations and anomalous morphological deviations. The minimum effective duration parameter in the length verification acts as a time-scale filter, filtering out anomalous signals with extremely short durations that may be caused by random noise or transient interference (such as vibrations from nearby vehicles), thereby improving the reliability of the anomaly detection results in an engineering sense.

[0042] Example 2: See Figure 3 In practice, the order of the polynomial regression fit is not fixed, but adaptively selected based on the number of data points and numerical distribution characteristics of each final outlier data segment. This selection process can be achieved by calculating the sum of squared residuals from polynomial fits of different orders, as shown in the following formula:

[0043] In conjunction with model selection criteria such as the Akaike Information Criterion (AIC), the AIC formula is as follows:

[0044] Where RSS represents the sum of squared residuals, This represents the i-th displacement measurement value actually collected. The value represents the i-th fitted value of the polynomial fit, n represents the number of displacement measurements in the data segment, AIC represents the value used to balance the goodness of fit and complexity of the model, k represents the number of parameters in the model, and L represents the likelihood function value of the model.

[0045] The goal is to obtain a smooth curve representing the long-term displacement trend of the final abnormal data segment, i.e., a trend curve. After fitting, the system calculates the difference between each fitted point on the trend curve and the actual collected displacement measurement value. This sequence of differences constitutes the residual sequence, which reflects the degree to which the actual observed value deviates from the trend line and contains short-term, local fluctuation information of the data segment. After obtaining the residual sequence, the trend assessment unit needs to extract peak and trough points from the residual sequence. In practice, the identification of peaks and troughs is accomplished by detecting the change in the sign of the first-order difference of the residual sequence. Specifically, when the first-order difference value of a data point changes from positive to negative, this point is marked as a peak point, and when the first-order difference value changes from negative to positive, this point is marked as a trough point. The specific formula for the first-order difference is:

[0046] in, This represents the first difference value of the i-th data point in the residual sequence. This represents the i-th residual value in the residual sequence. This represents the (i+1)th residual value in the residual sequence.

[0047] The system iterates through the entire residual sequence to identify all peaks and troughs, and calculates the vertical distance between each peak and its adjacent subsequent trough. The absolute value of this distance is recorded as a volatility value. The arithmetic mean of all such volatility values ​​is calculated, and this mean is defined as the local volatility of the final outlier segment. Local volatility quantifies the average strength of the residual sequence's oscillations around the trend curve. The specific formula for calculating the arithmetic mean of all such volatility values ​​is as follows:

[0048] in, This represents the local fluctuation amplitude of the final abnormal data segment. represents the vertical distance between the j-th peak point and the adjacent subsequent valley point, and m represents the number of all such fluctuation amplitude values.

[0049] After obtaining the local fluctuation amplitude of each final outlier data segment, the trend assessment unit begins to integrate the local fluctuation amplitudes of all final outlier data segments to obtain the overall trend fluctuation characteristics. The integration method uses a weighted average algorithm, and the specific formula is as follows:

[0050] in, Represents the overall trend and fluctuation characteristics. The weight coefficient represents the i-th final outlier data segment. represents the local fluctuation amplitude of the i-th final outlier data segment, and k represents the total number of final outlier data segments.

[0051] In the weighted average algorithm, the local fluctuation amplitude of each final abnormal data segment is assigned a weight coefficient. In some embodiments, the weight coefficient can be based on the number of data points contained in the final abnormal data segment; that is, the longer the data segment, the greater the weight of its corresponding local fluctuation amplitude in the weighted average calculation. This weighting method gives greater importance to abnormal events with longer durations. Optionally, the weight coefficient can also be determined comprehensively by combining the initial fluctuation intensity of the final abnormal data segment or other significance indicators. The result of the weighted average calculation is the overall trend fluctuation characteristic, which provides a single, comprehensive measure of the volatility of all abnormal segments throughout the entire monitoring period. The process of polynomial regression fitting typically uses the least squares method to solve for the polynomial coefficients. The least squares method finds the trend curve closest to the data points by minimizing the sum of squared residuals. In practice, this numerical calculation process may rely on robust matrix algorithms such as singular value decomposition to ensure the stability of the solution. For short or low-quality final outlier segments, the system may force the use of lower-order (e.g., linear or quadratic) polynomials for fitting to prevent overfitting. The quality of the trend curve directly affects the accuracy of the residual sequence, and thus the reliability of the calculation of local fluctuation amplitudes.

[0052] Extracting peaks and troughs from the residual sequence is fundamental to calculating local fluctuation amplitude. In practice, to improve the noise resistance of peak and trough detection, a mild smoothing filter can be applied to the residual sequence before first-order difference calculation, or peaks and troughs can be required to exceed a minimum amplitude threshold set based on the standard deviation of the residual sequence to be confirmed. When calculating the average distance between peaks and troughs, the system ensures that the pairing relationship is continuous, i.e., only the distance between a peak and its immediately following trough is calculated, or vice versa, to avoid amplitude calculation distortion caused by cross-period matching. Local fluctuation amplitude, as a statistic, effectively captures the instability and fluctuation degree within each final anomalous data segment. A weighted average algorithm is used to integrate local fluctuation amplitudes. Its core lies in the determination of weighting coefficients. The weighting coefficient allocation principle based on the data segment length reflects the importance of signal duration, as anomalous patterns with longer durations often have greater significance in engineering safety assessments. In addition to the data segment length, optionally, the weighting coefficients can also be correlated with the fluctuation intensity initially calculated by the sequence segmentation unit, so that data segments that were originally identified as having more severe fluctuations contribute more to the overall assessment. The overall trend fluctuation characteristics, as a scalar output, effectively summarize the overall system fluctuation state revealed by multiple abnormal data segments, providing quantitative input on trend stability for the monitoring and control unit to make decisions. The trend assessment unit transforms a series of local, discrete abnormal signals into a global, continuous trend stability assessment index.

[0053] Example 3: The matrix transformation unit receives continuous time series data from the data acquisition unit and organizes the time series data into a matrix structure. In specific implementation, the matrix transformation unit arranges the time series data into a Topplitz matrix. The construction of the Topplitz matrix depends on a preset hysteresis window length parameter. The construction process is as follows: Given a time series containing n displacement measurements... And a lag window length L (where 1 < L < n), the Topplitz matrix T is a matrix with A matrix with L rows and L columns, the i-th row of the Toplitz matrix T consists of L consecutive elements from the time series X, starting from index i. Each row of the Toplitz matrix T represents a subset of displacement sequences within a lag time window. Through this arrangement, the dynamic patterns in the time series are embedded into the linear structure of matrix T.

[0054] After successfully constructing the Topletz matrix T, the matrix transformation unit performs eigenvalue decomposition on T. In practice, eigenvalue decomposition is typically achieved by calculating the covariance matrix of the Topletz matrix T or by directly performing singular value decomposition on the Topletz matrix T. When using the covariance matrix approach, the covariance matrix of the Topletz matrix T is calculated. Then, eigenvalue decomposition is performed on the covariance matrix C, and the characteristic equation is solved. Thus, a set of eigenvalues ​​is obtained. The eigenvalues ​​and their corresponding eigenvectors v constitute the eigenvalue set. The eigenvalue set represents the energy distribution or variance contribution of the original time series data across different dimensions. After obtaining the eigenvalue set, the matrix transformation unit sorts all eigenvalues ​​in descending order of value, generating an ordered eigenvalue sequence. .

[0055] The matrix transformation unit then calculates the ratio sequence between consecutive eigenvalues ​​in the ordered eigenvalue sequence. The calculation is defined as the ratio of two adjacent eigenvalues, that is, for i from 1 to L-1, the calculation is... . ratio series This reflects the steepness of the eigenvalue decay; a larger ratio value indicates a significant drop in the eigenvalue at that point. The system then identifies the ratio sequence. Jump points in the sequence can be identified by finding ratio sequences. This can be achieved by finding local maxima, or by setting a dynamic threshold, which is used when a certain ratio value is reached. If the threshold is exceeded, position k is considered a jump point. Once the jump point position k is identified, all feature values ​​in the ordered feature value sequence that precede jump point k are... The eigenvalues ​​were identified as the main eigenvalue set, while the eigenvalues ​​after the jump point k were... Then it naturally belongs to the secondary eigenvalue group.

[0056] The feature derivation unit receives the eigenvalue set and the partitioning information of the primary and secondary eigenvalue groups from the matrix transformation unit. The primary task of the feature derivation unit is to calculate the energy concentration index. The energy concentration index E is defined as the ratio of the cumulative sum of all eigenvalues ​​in the primary eigenvalue group to the sum of the entire eigenvalue set. Its calculation formula is as follows:

[0057] Where E represents the energy percentage or a certain proportion indicator. k represents the number of the first k elements selected. Representing the The eigenvalues ​​(or energy values, etc.) corresponding to each element. The index represents the sort index of the element, and L represents the total number of all elements.

[0058] Energy concentration E is a value between 0 and 1 that quantifies the proportion of the main variation patterns in the data explained by the first k principal components. A higher energy concentration E value indicates a stronger structure in the time series and a relatively lower component of random noise. In some embodiments, to enhance numerical stability, the sum in the denominator can be calculated only for those eigenvalues ​​greater than a small positive number to avoid the influence of numerical calculation errors.

[0059] The feature derivation unit calculates the feature dispersion index D, which measures the relative difference between the variance within the principal eigenvalue group and the variance within the secondary eigenvalue group. The calculation process consists of two steps: calculating the variances within the principal eigenvalue group separately... variance and secondary eigenvalue groups variance The characteristic dispersion index D is calculated using the following formula:

[0060] Where D represents the difference or discrimination index. The variance representing the principal components (or the main dataset). Represents the variance of the minor components (or minor datasets).

[0061] The characteristic dispersion index D ranges from -1 to 1. When the variance within the principal eigenvalue group is much greater than that within the secondary eigenvalue group (i.e., the principal components have concentrated energy and the eigenvalues ​​decay rapidly), the characteristic dispersion index D approaches 1; when the variances of the two groups are similar, the characteristic dispersion index D approaches 0; in extreme cases, if the variance of the secondary eigenvalue group is even larger, the characteristic dispersion index D is negative. The characteristic dispersion index D reveals the uniformity of the eigenvalue distribution from another dimension and is an effective supplement to the energy concentration index E.

[0062] The feature derivation unit ultimately combines the energy concentration E and the feature dispersion index D to derive the feature parameters used for the monitoring and control unit. In some embodiments, the derived feature parameters can be a two-dimensional vector. The two indicators are directly transmitted to the monitoring and control unit. Optionally, the characteristic parameter can also be a scalar, for example, through a linear combination. Generate, where weight coefficients and The settings are pre-defined based on the degree of concern regarding the concentration and dispersion of signal energy in the specific application scenario. In some embodiments, a product form can also be used. This process generates characteristic parameters, emphasizing that both energy concentration E and characteristic dispersion index D need to be at high levels simultaneously to produce meaningful characteristic parameter values. After deriving the characteristic parameters, the characteristic derivation unit outputs them to the monitoring and control unit, providing a data foundation for dynamically adjusting the monitoring strategy. The combined work of the matrix transformation unit and the characteristic derivation unit transforms the macroscopic characteristics of the one-dimensional time series into refined characteristic parameters with clear physical meaning.

[0063] See Figure 4 This figure focuses on the characteristic derivative units of the real-time surface monitoring system for tunnel boring machine (TBM) construction. Using time step as the horizontal axis, it synchronously presents the dynamic changes in energy concentration and characteristic dispersion indices. Energy concentration quantifies the proportion of data variation explained by the main characteristic value group; a higher value indicates stronger data structure and a lower proportion of random noise. The characteristic dispersion index reflects the variance difference between the main and secondary characteristic value groups, demonstrating the uniformity of characteristic value distribution. This figure visually presents the intrinsic characteristic evolution of surface displacement data during TBM construction, providing a quantitative basis for the monitoring and control unit to dynamically adjust sampling frequency, report generation interval, and alarm strategies. It helps engineers accurately identify the correlation between data quality and ground stability, effectively supporting the safety management of TBM construction.

[0064] Example 4: The monitoring and control unit performs a multiplication operation on the energy concentration and characteristic dispersion indicators. The result is defined as a signal reliability score, a dimensionless scalar value used to comprehensively evaluate the overall quality and reliability level of the current monitoring data. The monitoring and control unit internally establishes a mapping relationship between the signal reliability score and monitoring strategy parameters, dynamically adjusting the system's data sampling frequency and settlement report generation interval based on the real-time calculated signal reliability score. When the signal reliability score remains in a high range, the system determines that the current data is stable and reliable, and can appropriately reduce the sampling frequency of the data acquisition unit and extend the report generation interval to save system resources and reduce data transmission volume. When the signal reliability score falls below a preset safety threshold, the monitoring and control unit immediately triggers a high-frequency monitoring mode. High-frequency monitoring mode means that the system will instruct the data acquisition unit to collect displacement data at the highest available frequency. Simultaneously, the monitoring and control unit will generate and send real-time alarm information to the monitoring center, reminding staff to pay attention to potential surface subsidence risks.

[0065] The data acquisition unit, as a fundamental component of the system, is based on a large number of surface displacement sensors deployed throughout the monitoring area. In practice, these sensors are arranged in a grid according to pre-planned geographical coordinates. The grid density and shape are determined based on the tunnel axis, the expected impact range, and the location of key protected structures. The data acquisition unit periodically collects raw readings from all surface displacement sensors according to instructions from the monitoring and control unit. The collected raw data typically contains some environmental noise and random fluctuations. To improve data quality, the data acquisition unit applies digital filters to smooth the raw data. These filters can be either finite-length unit impulse response (FIR) filters or infinite-length unit impulse response (IRS) filters, with parameters configured according to signal characteristics. During data transmission or acquisition, data points may be missing due to temporary sensor failure or communication interruptions. The data acquisition unit uses linear interpolation to compensate for these missing points. Linear interpolation calculates estimates using adjacent valid data points before and after the missing point, thus outputting a continuous and complete time-series data stream for subsequent sequence segmentation units. The specific linear interpolation formula is as follows:

[0066] in, This represents the displacement estimate at the missing time t. This represents the effective displacement measurement value adjacent to the missing time t. This represents the time corresponding to the valid measurement value. This represents the effective displacement measurement value adjacent to the missing time t. This represents the time corresponding to the valid measurement value, and satisfies... .

[0067] The specific application logic of the signal reliability score can be implemented through a pre-defined lookup table, which clarifies the specific operational parameters corresponding to different score ranges. Refer to Table 1, which illustrates a possible correspondence between signal reliability scores and monitoring strategies.

[0068] Table 1: Mapping Table of Signal Reliability Score and Monitoring Strategy

[0069] The reference frequency and reference interval are default parameters used by the system under normal operating conditions, preset by engineers according to monitoring specifications. Setting the safety threshold is crucial, typically around 0.4. When the signal reliability score falls below this threshold, it indicates low energy concentration or abnormally dispersed feature distribution, potentially signifying strong interference with the monitoring system or abnormal surface deformation. After triggering the high-frequency monitoring mode, the monitoring control unit continuously monitors changes in the signal reliability score. Once the score recovers above the safety threshold and stabilizes for a period, the system automatically reverts to the normal monitoring mode.

[0070] The implementation details of the data acquisition unit also include sensor selection and calibration. Surface displacement sensors can be hydrostatic levels, total stations, or GPS receivers. In some embodiments, multiple types of sensors may be deployed together for mutual verification. The data acquisition unit typically consists of a main control module and multiple data acquisition nodes. The main control module receives instructions from the monitoring and control unit and distributes them to each node, while also aggregating the data uploaded by each node. The design of digital filters needs to balance smoothing effects with signal fidelity. Optionally, moving average filters can be used for fast and simple smoothing, or frequency domain filters based on Fourier transform can be used to remove noise at specific frequencies. While linear interpolation is simple, it is suitable for most cases where data loss is not severe. For cases with long, continuous missing data segments, the data acquisition unit may mark these segments as less reliable for subsequent units to reference. The continuous time series data ultimately output by the data acquisition unit is the foundation for all subsequent advanced analyses; its quality directly affects the accuracy of the signal confidence score, and thus the decision-making efficiency of the entire system.

[0071] See Figure 5 This figure focuses on the core functions of the monitoring and control unit, using time step as the horizontal axis to synchronously present the dynamic changes of signal reliability score and relative sampling frequency, with safety thresholds marked by dashed lines. The signal reliability score is obtained by multiplying energy concentration and characteristic dispersion index, quantifying the overall quality and reliability of the monitoring data, while the relative sampling frequency reflects the density of data collection. The figure clearly demonstrates the mechanism by which the monitoring and control unit dynamically adjusts its strategy based on data quality: when data reliability is insufficient, monitoring is strengthened by increasing the sampling frequency, ensuring timely identification of surface subsidence risks during tunnel boring machine construction, and fully demonstrating the system's intelligent control and on-demand monitoring advantages.

[0072] Example 5: The data calibration unit is used to periodically calibrate sensor data against a reference point. In specific implementation, the reference point is a high-precision measurement marker set on a stable stratum outside the monitoring area, whose coordinates are considered constant or negligible through long-term observation. The data calibration unit automatically initiates a calibration process according to a preset calibration cycle, such as every 24 hours. During calibration, the data calibration unit synchronously reads the measurement values ​​of all surface displacement sensors at the current moment and simultaneously acquires the coordinate data of each reference point at the same time. The data calibration unit calculates the difference between the measurement value of each surface displacement sensor and the coordinate data of the corresponding reference point at the geographical location. This difference is defined as the systematic error, which reflects the long-term measurement deviation caused by factors such as temperature drift, mechanical stress, or loose installation. After acquiring the systematic error, the data calibration unit applies an error compensation algorithm to adjust subsequent displacement measurements. The error compensation algorithm can use a simple offset correction model, that is, subtracting the calculated systematic error value from each original displacement measurement value. The specific calculation formula is as follows:

[0073] in, This represents the displacement measurement value after offset correction (calibrated data). This represents the original displacement measurement value collected by the sensor at time t. The system error (i.e., the difference between the sensor measurement and the reference point data) is calculated by the data calibration unit.

[0074] If a systematic error is found to have a strong correlation with time or ambient temperature, the error compensation algorithm can also use a linear regression model to establish a functional relationship between the systematic error and time or temperature using historical calibration data, thereby providing more accurate dynamic compensation for the measured values. Taking ambient temperature as an example, the specific formula for the linear regression relationship between the systematic error and temperature is as follows:

[0075] The corresponding dynamic compensation formula is as follows:

[0076] in, denoted by T, represents the systematic error at ambient temperature T, k represents the slope of the linear regression model (the coefficient of systematic error as a function of temperature), and b represents the intercept of the linear regression model (the baseline systematic error at temperature 0). Representing time t and ambient temperature as Displacement measurement value after dynamic compensation The ambient temperature measurement value corresponding to time t.

[0077] The displacement measurement value, adjusted by the error compensation algorithm, is output to the data acquisition unit to generate subsequent time series data. This process significantly reduces the impact of the sensor's own systematic bias on the monitoring results.

[0078] The monitoring and control unit generates a visual report containing the location of abnormal data segments and trend assessment results. In practice, this report is presented in a graphical interface combined with data tables. The location information of abnormal data segments comes from the processing results of the sequence segmentation unit and the anomaly marking unit. In the report, this information is typically overlaid on the time-displacement curve as a highlighted line segment, with the start time, duration, and maximum displacement of the anomaly marked. The trend assessment results come from the trend assessment unit and are usually displayed as a fitted trend curve and marked overall trend fluctuation characteristic values. The report also integrates feature parameters provided by the matrix transformation unit and the feature derivation unit, such as signal confidence scores, providing users with a comprehensive view of the data health status. When the feature dispersion index rises abnormally, the monitoring and control unit automatically activates the redundant sensor network for cross-validation. The judgment of an abnormal rise in the feature dispersion index is based on comparing it with a dynamic threshold obtained from historical data statistics. For example, the system may calculate the moving average and standard deviation of the feature dispersion index over the past 7 days. When the current feature dispersion index value exceeds the moving average plus three times the standard deviation, it is determined to be an abnormal rise. Once the trigger condition is met, the monitoring and control unit sends an activation command to the redundant sensor network. A redundant sensor network consists of backup sensors deployed near the main sensor but on different lines or using different measurement principles. Once activated, these redundant sensors begin to collect data synchronously.

[0079] The monitoring and control unit performs cross-validation using data collected by a redundant sensor network. The cross-validation process involves comparing the measurement results from the primary sensor and redundant sensors at the same location points, calculating consistency indices such as correlation coefficients or mean absolute error. If the cross-validation results show a significant deviation in the primary sensor data, the monitoring and control unit will mark that sensor data as unreliable in the generated detailed analysis results and recommend that the redundant sensor data be used as the standard. The detailed analysis results are output to the monitoring platform, including the values ​​of anomaly dispersion indicators, the time when cross-validation was triggered, the list of sensors participating in the cross-validation, the calculation results of the consistency indices, and a final conclusion regarding data reliability. This mechanism greatly enhances the system's robustness and reliability in the face of individual sensor failures or localized interference.

[0080] The workflow of the data calibration unit can be illustrated using a specific calibration cycle as an example. In practice, assume the calibration cycle is set to 2:00 AM daily, when tunnel boring machine construction is typically paused and environmental interference is minimal. The data calibration unit first sends instructions to all sensors and reference point measuring equipment (such as a total station) to acquire data at the same timestamp. For a hydrostatic level sensor numbered S01, its reading is +15.3 mm, while the corresponding reference point's precise elevation change measurement at this moment is +2.1 mm. Therefore, the systematic error calculated by the data calibration unit is... Millimeters. Over the next 24 hours, all raw displacement measurements from this S01 sensor will be automatically subtracted by a 13.2 mm system error value before being used in subsequent analysis. Optionally, the data calibration unit will record the system error calculated from each calibration. If the system error of a sensor shows a linear increasing trend over time, it may trigger an alert indicating that the sensor needs maintenance or replacement.

[0081] The specific content and format of the visualization report generated by the monitoring and control unit may vary in some embodiments. In some cases, the report may be an interactive webpage, allowing users to select different time ranges for viewing. The report's homepage displays a map of the entire monitoring area, with different colors used to mark the current status of each monitoring point (e.g., Normal, Caution, Alarm). Clicking on a specific monitoring point opens a detailed window displaying the point's displacement time series curve. Abnormal data segments are highlighted with a red background, and a polynomial trend curve obtained from the trend assessment unit is plotted using dashed lines. A separate section of the report lists detailed information for all abnormal data segments in tabular form, including the abnormal data segment number, center time, duration, maximum displacement within the segment, local fluctuation amplitude, and overall trend fluctuation characteristics. Another part of the report displays system-level characteristic parameters, such as the most recently calculated energy concentration, characteristic dispersion index, and signal reliability score, visually displaying their values ​​and corresponding level ranges in a dashboard format.

[0082] A specific scenario for activating the redundant sensor network when the characteristic dispersion index rises abnormally can be exemplified by assuming a critical phase where the tunnel crosses an underground river. If the characteristic dispersion index calculated by the monitoring and control unit exceeds the historical upper limit of the normal range for several consecutive cycles, the system determines that the characteristic dispersion index has risen abnormally. The monitoring and control unit then activates the redundant sensor network deployed on both sides of the axis. These redundant sensors may employ laser ranging principles, complementing the hydrostatic level on the main line. After the activation command is issued, the redundant sensor network collects data at a higher frequency for a predetermined period. The monitoring and control unit compares the data from the main line sensors and the redundant sensors in real time during this period, calculating the difference between each set of synchronized data. If the comparison reveals that most redundant sensors show stability, while the main line sensor data fluctuates drastically and differs significantly from the redundant data, detailed analysis will indicate that the main line sensor data may be subject to unexplained interference. Engineers are advised to focus on the data from the redundant sensor network and to inspect the main line sensors.

[0083] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A real-time surface monitoring system for shield tunnel construction, characterized in that, The system includes a data acquisition unit, a sequence segmentation unit, an anomaly marking unit, a trend evaluation unit, a matrix transformation unit, a feature derivation unit, and a monitoring and control unit; The data acquisition unit is configured to acquire displacement measurement values ​​from surface sensors and generate time series data; the sequence segmentation unit is configured to analyze abrupt changes in the time series data and divide the sequence into multiple data segments based on the abrupt changes; the anomaly marking unit is configured to perform statistical tests on each data segment to identify a set of anomalous data segments. The trend assessment unit is configured to fit a trend line to the set of abnormal data segments and assess the degree of fluctuation of the fitting deviation; the matrix transformation unit is configured to organize the time series data into a specific matrix structure and perform matrix decomposition; the feature derivation unit is configured to derive feature parameters from the matrix decomposition results. The monitoring and control unit is configured to adjust the monitoring strategy and generate a settlement report based on characteristic parameters.

2. The real-time surface monitoring system based on shield tunnel construction as described in claim 1, characterized in that, The sequence segmentation unit uses a sliding window technique to scan the time series data and calculates the variance of the displacement values ​​within each window. When the rate of change of variance between consecutive windows exceeds a preset threshold, the position is recorded as a mutation point. The time series data is divided into independent data segments from each mutation point. The displacement range and standard deviation of each data segment are calculated, and the product of the range and standard deviation is used as the fluctuation intensity of the data segment. A grouping algorithm is used to cluster the fluctuation intensity of all data segments, and the data segment corresponding to the group with the largest fluctuation intensity is selected as a candidate abnormal data segment.

3. The real-time surface monitoring system based on shield tunnel construction as described in claim 2, characterized in that, The anomaly marking unit calculates the similarity measure between each candidate anomaly data segment and its preceding and following adjacent data segments. The similarity measure is based on the distribution pattern of displacement values ​​in the data segment. A similarity threshold is set, and data segments with similarity below the threshold are initially marked as suspicious anomaly segments. The length of the suspicious anomaly segments is verified, and only data segments with a length greater than the minimum valid duration are retained as the final set of anomaly data segments.

4. The real-time surface monitoring system based on shield tunnel construction as described in claim 3, characterized in that, The trend assessment unit performs polynomial regression fitting on each final abnormal data segment to obtain a trend curve; calculates the residual sequence between the trend curve and the actual displacement value; extracts the peak and valley values ​​of the residual sequence and calculates the average distance between the peak and valley values ​​as the local fluctuation amplitude; integrates the local fluctuation amplitudes of all final abnormal data segments and obtains the overall trend fluctuation characteristics by weighted averaging.

5. The real-time surface monitoring system based on shield tunnel construction as described in claim 1, characterized in that, The matrix transformation unit arranges the time series data into a Topplitz matrix, where each row of the Topplitz matrix represents the displacement sequence of a lag time window; performs eigenvalue decomposition on the Topplitz matrix to obtain an eigenvalue set; arranges the eigenvalues ​​in descending order and calculates the ratio sequence between consecutive eigenvalues; identifies jump points in the ratio sequence and determines the eigenvalues ​​before the jump points as the main eigenvalue group.

6. The real-time surface monitoring system based on shield tunnel construction as described in claim 5, characterized in that, The feature derivation unit separates the primary feature value group and the secondary feature value group from the feature value set; the ratio of the cumulative sum of the primary feature value group to the sum of the entire feature value set is calculated as the energy concentration. The relative difference between the variance of the primary eigenvalue group and the variance of the secondary eigenvalue group is calculated as the eigenvalue dispersion index. Characteristic parameters are derived by combining energy concentration and characteristic dispersion indices.

7. The real-time surface monitoring system based on shield tunnel construction as described in claim 6, characterized in that, The monitoring and control unit multiplies the energy concentration and characteristic dispersion index to obtain the signal reliability score; it dynamically adjusts the data sampling frequency and report generation interval based on the signal reliability score; when the signal reliability score is lower than the safety threshold, it triggers the high-frequency monitoring mode and sends a real-time alarm.

8. The real-time surface monitoring system based on shield tunnel construction as described in claim 1, characterized in that, The data acquisition unit includes multiple surface displacement sensors, which are arranged in a grid according to geographic coordinates. The data acquisition unit collects sensor data at regular intervals and applies digital filters to smooth data fluctuations. Linear interpolation is used to compensate for missing data points and output continuous time series data.

9. The real-time surface monitoring system based on shield tunnel construction as described in claim 1, characterized in that, The system also includes a data calibration unit for periodically calibrating sensor data against a reference point; the data calibration unit calculates the systematic error between the sensor data and the reference point, and applies an error compensation algorithm to adjust the displacement measurement value.

10. The real-time surface monitoring system based on shield tunnel construction as described in claim 7, characterized in that, The monitoring and control unit generates a visual report containing the location of abnormal data segments and trend assessment results; when the characteristic dispersion index rises abnormally, it automatically activates the redundant sensor network for cross-validation and outputs detailed analysis results to the monitoring platform.

Images