A health monitoring system for comprehensive pipe gallery foundation pit support structure

Abstract

The present invention relates to the technical field of foundation pit support structure monitoring, and discloses a health monitoring system for foundation pit support structures of an integrated pipeline corridor, comprising: a monitoring unit, an excavation unit, an analysis unit, and a determination unit, wherein the monitoring unit is used to divide the monitoring target object into three monitoring areas, collect heterogeneous monitoring data of each monitoring area in real time, and generate a pre-processed original data set. The monitoring target object is the support structure of the foundation pit in a soft soil area. The present invention divides the monitoring object into three areas, collects heterogeneous data in a targeted manner, mines creep characteristics, generates a regionalized time-varying threshold, solves the problems of fixed thresholds and noise masking, and can promptly detect early minute progressive deformations caused by soft soil creep. Through progressive damage judgment, the system combines regional damage deviation with cross-regional correlation values ​​to locate damage and provide graded warnings, and can also trigger collaborative warnings, thereby improving monitoring accuracy, positioning accuracy, and warning timeliness.

Description

Technical Field

[0001] The present invention relates to the technical field of foundation pit support structure monitoring, and in particular to a health monitoring system for a foundation pit support structure of an integrated pipe gallery. Background Art

[0002] The health monitoring of the foundation pit support structure of the integrated pipeline corridor mainly relies on the coordinated work of multiple types of sensors. By deploying multiple sensors at key locations of the support structure, the parameters of the support structure are collected in real time. These data are aggregated to the monitoring center via wired or wireless transmission. The system independently compares and analyzes single parameters based on preset static thresholds, and triggers an audible and visual alarm when a parameter exceeds the threshold.

[0003] However, in practical applications, especially in soft soil areas, soft soil creep often causes early damage to support structures. Early damage is manifested as the progressive accumulation of small strains. Such changes usually develop slowly at a very slow deformation rate. Existing monitoring systems overly rely on fixed thresholds and tend to ignore such features in noise. As a result, such low-amplitude, long-period progressive deformations are masked by background noise and cannot be discovered in time, affecting the health monitoring effect of the support structure. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides a health monitoring system for a comprehensive pipe gallery foundation pit support structure to solve the above problems.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions:

[0006] A comprehensive pipe gallery foundation pit support structure health monitoring system, comprising:

[0007] The monitoring unit is used to divide the monitoring target object into three monitoring areas, collect heterogeneous monitoring data from each monitoring area in real time, and generate a pre-processed raw data set. The monitoring target object is the support structure of the foundation pit in the soft soil area;

[0008] A mining unit is used to perform feature mining on the original data set and generate a creep feature vector set;

[0009] An analysis unit, used for analyzing the creep characteristic vector set to obtain a monitoring regionalized time-varying threshold set;

[0010] The judgment unit is used to perform progressive damage judgment on the creep characteristic vector set and the monitoring regional time-varying threshold set, locate the damage position of the monitoring target object and issue an early warning.

[0011] Further, the three monitoring areas include:

[0012] Edge stress zone, middle stable zone and corner stress concentration zone;

[0013] Among them, the edge stress zone is the edge where the support structure contacts the surrounding soil;

[0014] The middle stable zone is the middle part of the support structure;

[0015] The corner stress concentration area is the corner of the foundation pit plane.

[0016] Furthermore, feature mining is performed on the original data set to generate a creep feature vector set, including:

[0017] The original data set is decomposed into three monitoring areas to obtain wavelet energy entropy parameters, stress mutation sensitivity factors and time-varying fractal stability parameters, forming a regional characteristic parameter group;

[0018] Based on the regional characteristic parameter group, the temporal similarity parameters between each monitoring area are calculated;

[0019] Decompose the time series similarity parameters and construct a feature matrix reflecting the time series correlation of the monitoring area;

[0020] Reconstruct the phase space of the feature matrix to generate a time-varying threshold boundary;

[0021] The regional feature parameter group, temporal similarity parameter and time-varying threshold boundary are fused to form a creep feature vector set.

[0022] Furthermore, the original data set is decomposed into three monitoring areas to obtain wavelet energy entropy parameters, stress mutation sensitivity factors and time-varying fractal stability parameters, forming a regional characteristic parameter group, including:

[0023] Perform noise separation on the original data set to obtain the noise energy ratio parameter;

[0024] Decompose the original data set according to the noise energy ratio parameter to obtain the energy weight parameter of each frequency band and form the weighted initial feature set;

[0025] The weighted initial feature set is calculated to obtain the wavelet energy entropy parameters of the edge stress area;

[0026] The weighted initial feature set is decomposed and adjusted to obtain the stress mutation sensitivity factor in the central stable zone;

[0027] The fractal dimension of the weighted initial feature set is calculated to obtain the time-varying fractal stability parameters of the corner stress concentration zone.

[0028] The wavelet energy entropy parameter, stress mutation sensitivity factor and time-varying fractal stability parameter are normalized and then weighted fused to generate a regional characteristic parameter group.

[0029] Furthermore, the creep eigenvector set is analyzed to obtain a set of monitoring regionalized time-varying thresholds, including:

[0030] Set the risk factor for each monitoring area;

[0031] Based on the risk coefficient of each monitoring area, the creep characteristic vector of the monitoring area is weighted and fused to generate a regional coupling weight matrix that reflects the coupling effect of the characteristics of different monitoring areas.

[0032] Based on the regional coupling weight matrix, the creep eigenvector set is statistically analyzed to generate a dynamic threshold benchmark curve.

[0033] The dynamic threshold benchmark curve is modified according to the time series similarity parameter to obtain the monitoring regional time-varying threshold set.

[0034] Furthermore, based on the regional coupling weight matrix, the creep eigenvector set is statistically analyzed to generate a dynamic threshold benchmark curve, including:

[0035] The creep eigenvector set is calculated by an adaptive sliding window algorithm to generate a two-parameter eigenvector;

[0036] The double-parameter eigenvector is analyzed based on the regional coupling weight matrix to obtain the soft soil creep stage identification factor.

[0037] The dual-parameter eigenvector is mapped according to the soft soil creep stage identification factor to generate a dynamic threshold benchmark curve.

[0038] Furthermore, a progressive damage determination is performed on the creep feature vector set and the monitoring regionalized time-varying threshold set, and the damage position of the monitoring target object is located and an early warning is issued, including:

[0039] Based on the risk coefficient, the creep characteristic vector set and the monitoring regional time-varying threshold set are regionally operated to generate the regional damage deviation.

[0040] The regional damage deviation is corrected according to the time series similarity parameter to obtain the cross-regional damage correlation value;

[0041] The regional damage deviation and cross-regional damage correlation values ​​are weighted and integrated to generate a comprehensive damage location index;

[0042] When the comprehensive damage location index exceeds the regional time-varying threshold of the corresponding monitoring area, a graded warning is triggered for the monitoring area, and the damage location is located by the index peak.

[0043] Furthermore, locating the damage location by index peak includes:

[0044] Perform spatial discretization on the comprehensive damage location index to generate the local peak coefficient of each monitoring area;

[0045] The local peak coefficient is calculated according to the risk factor of each monitoring area to obtain the weighted peak response value;

[0046] The extreme point position of the weighted peak response value is analyzed to generate the location and confidence parameters of the damage position.

[0047] Furthermore, when the comprehensive damage location index exceeds the regional time-varying threshold of the corresponding monitoring area, a graded warning is triggered for the monitoring area, including:

[0048] The comprehensive damage location index in the corner stress concentration area 1.0 and continuing If the time is more than one hour, a level one warning will be triggered;

[0049] Comprehensive damage location index in the edge stress zone 1.1 and continuing If the time is more than one hour, a second-level warning will be triggered;

[0050] Comprehensive damage localization index in the central stable zone 1.2 and continue Hours or more, triggering the third-level warning;

[0051] Among them, the severity of warning is level one warning > level two warning > level three warning;

[0052] If the cross-regional damage correlation values ​​of adjacent monitoring areas are all greater than 0.7, a coordinated warning will be triggered.

[0053] Furthermore, if the cross-regional damage correlation values ​​of adjacent monitoring areas are all greater than 0.7, a coordinated warning is triggered, including:

[0054] If the cross-regional damage correlation values ​​of any two adjacent monitoring areas in the edge stress zone, corner stress concentration zone and central stable zone are all greater than 0.7, a coordinated level 1 warning will be triggered;

[0055] If the cross-regional damage correlation values ​​of the edge stress zone, corner stress concentration zone, and central stable zone are all greater than 0.7, a coordinated secondary warning is triggered;

[0056] Among them, the severity of the warning is coordinated level 2 warning > coordinated level 1 warning.

[0057] In summary, the present invention mainly has the following beneficial effects:

[0058] By dividing the support structure into three monitoring areas: the edge stress zone, the middle stable zone, and the corner stress concentration zone, differentiated monitoring is carried out on the stress characteristics of different areas, solving the problem of insufficient sensitivity to local characteristics caused by traditional global unified monitoring. It also integrates multi-dimensional parameters such as wavelet energy entropy, stress mutation sensitivity factor, time-varying fractal stability parameter, and introduces time series similarity analysis for cross-regional data correlation, which can effectively extract the small strain accumulation characteristics caused by soft soil creep from the noise. For example, by separating the noise energy ratio through wavelet energy entropy and combining it with the weighted initial feature set decomposition, it can accurately capture the progressive deformation trend of the edge stress zone, and the fractal dimension calculation can reveal the complex evolution law of deformation in the corner area, avoid low-amplitude and long-period deformation being masked by background noise, and achieve low missed reports and high-sensitivity rapid identification of early damage.

[0059] Through regional risk coefficient weighting, coupling weight matrix analysis and adaptive sliding window algorithm, a dynamic threshold benchmark curve was constructed, and combined with real-time correction of time series similarity parameters, a regionalized time-varying threshold set was formed. Specifically, differentiated weights were assigned according to the risk level of each region, and the nonlinear evolution boundary of the deformation sequence was characterized by phase space reconstruction technology, so that the threshold could be dynamically adjusted with the soft soil creep stage (such as attenuated creep, stable creep, and accelerated creep). In the early stage of soft soil creep, the system dynamically reduced the threshold sensitivity through dual-parameter eigenvectors and stage identification factors to avoid false alarms. As deformation accumulates, the threshold boundary is gradually tightened to ensure timely warning in the accelerated deformation stage. This adaptive and regionalized threshold mechanism not only overcomes the lag of fixed thresholds to environmental changes, but also avoids the problem of misjudgment of single parameters, thereby improving the reliability of the monitoring system in complex soft soil environments.

[0060] Through the calculation of regional damage deviation, cross-regional correlation correction and fusion of comprehensive damage location index, a progressive judgment system of regional local analysis-cross-regional correlation verification-global positioning was constructed. On the one hand, the regional damage deviation based on risk factor weighting can quickly lock the abnormal area. On the other hand, the cross-regional damage correlation value can identify the deformation chain effect, trigger collaborative early warning, and avoid the evolution of single regional damage into overall instability. Through spatial discretization processing and weighted peak response analysis, the damage location accuracy can be improved to the characteristic points of the specific monitoring area, combined with the confidence parameter to provide a quantitative basis for engineering disposal. This hierarchical early warning mechanism not only achieves accurate quantification of the damage degree, but also improves the systematic health monitoring of foundation pit support structures and the efficiency of emergency disposal through the cross-regional linkage response mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of the health monitoring system of the comprehensive pipeline corridor foundation pit support structure of the present invention. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0063] refer to Figure 1 , a comprehensive pipe gallery foundation pit support structure health monitoring system, including:

[0064] The monitoring unit is used to divide the monitoring target object into three monitoring areas, collect heterogeneous monitoring data from each monitoring area in real time, and generate a pre-processed raw data set. The monitoring target object is the support structure of the foundation pit in the soft soil area;

[0065] A mining unit is used to perform feature mining on the original data set and generate a creep feature vector set;

[0066] An analysis unit, used for analyzing the creep characteristic vector set to obtain a monitoring regionalized time-varying threshold set;

[0067] The judgment unit is used to perform progressive damage judgment on the creep characteristic vector set and the monitoring regional time-varying threshold set, locate the damage position of the monitoring target object and issue an early warning.

[0068] Through the zoning monitoring mechanism, the support structure is divided into three major monitoring areas, realizing differentiated monitoring of areas with different geological conditions and stress characteristics, solving the monitoring problems of nonlinear deformation laws and large temporal and spatial differences in soft soil foundation pits, making data collection more targeted and comprehensive. Heterogeneous data fusion preprocessing technology effectively integrates multi-dimensional monitoring data such as displacement, stress, and moisture content, providing a high-precision raw data set for subsequent analysis and avoiding the one-sidedness of single indicator monitoring.

[0069] The time-varying threshold analysis module based on creep feature mining can dynamically capture the progressive damage characteristics of the support structure caused by the creep effect of soft soil. Through the designed monitoring regional time-varying threshold, it breaks through the hysteresis defect of traditional fixed threshold monitoring and can reflect the performance evolution law of the foundation pit support structure under the action of time in real time. The progressive damage judgment mechanism combined with the spatial positioning algorithm realizes a three-level diagnosis from feature recognition, threshold warning to precise positioning. It can not only predict the structural damage trend in advance, but also accurately lock the damage location of key parts such as retaining piles and support systems, providing targeted decision-making basis for engineering rescue, and significantly improving the health monitoring efficiency of foundation pit support structures in soft soil areas.

[0070] In one case of this embodiment, the three monitoring areas include:

[0071] Edge stress zone, middle stable zone and corner stress concentration zone;

[0072] Among them, the edge stress zone is the edge where the support structure contacts the surrounding soil;

[0073] The middle stable zone is the middle part of the support structure;

[0074] The corner stress concentration area is the corner of the foundation pit plane.

[0075] In one case of this embodiment, feature mining is performed on the original data set to generate a creep feature vector set, including:

[0076] The original data set is decomposed into three monitoring areas to obtain wavelet energy entropy parameters, stress mutation sensitivity factors and time-varying fractal stability parameters, forming a regional characteristic parameter group;

[0077] Based on the regional characteristic parameter group, the time series similarity parameters between each monitoring area are calculated, specifically including: performing range normalization processing on the wavelet energy entropy parameter, stress mutation sensitivity factor and time-varying fractal stability parameter; extracting the time series data series of the three parameters for each monitoring area respectively; using the dynamic time warping algorithm to align the time axis of each parameter series; calculating the minimum cumulative distance between the edge stress area and the central stable area, the edge stress area and the corner stress concentration area, and the central stable area and the corner stress concentration area; calculating the linear correlation coefficient of each parameter series by the Pearson correlation coefficient; normalizing the minimum cumulative distance and weighted fusion with the correlation coefficient to form the time series similarity parameter between each monitoring area;

[0078] The time series similarity parameters were decomposed to construct a feature matrix reflecting the time series correlation of the monitoring areas. Specifically, the time series similarity parameters of the three groups of regions were decomposed into three independent dimensions according to the parameter type: wavelet energy entropy similarity, stress mutation sensitivity factor similarity, and time-varying fractal stability similarity. For each parameter dimension, a hierarchical clustering algorithm was used to perform cluster analysis on the similarity values ​​between regions to form energy entropy clustering clusters, stress mutation clustering clusters, and fractal stability clustering clusters. The clustering quality was evaluated by the silhouette coefficient and weights were assigned (energy entropy weight was 0.4, stress mutation weight was 0.3, and fractal stability weight was 0.3). The clusters of the three dimensions were clustered. The results were weighted and fused, and a 3×3 feature matrix was constructed with regional pairs as units: the matrix elements were generated by multiplying the three parameter similarity values ​​of the corresponding regional pairs by weights and then superimposing them. For example, the correlation strength between the edge stress zone and the central stable zone was obtained by summing the wavelet energy entropy similarity × 0.4, stress mutation similarity × 0.3, and fractal stability similarity × 0.3 of the regional pair. The main diagonal elements were directly assigned by the mean of the wavelet energy entropy of the regional parameter sequence itself, ultimately forming a three-dimensional correlation matrix that includes energy transfer correlation, stress synchronization characteristics, and deformation mode coordination. This correlation matrix can fully reflect the correlation relationship of the monitoring area time series.

[0079] The phase space of the feature matrix is ​​reconstructed to generate a time-varying threshold boundary. Specifically, the following steps are performed: by gradually increasing the dimension and calculating the rate of change of the Euclidean distance between adjacent data points (data points refer to the correlation strength values ​​between regions at each time step in the feature matrix, including three parameters: wavelet energy entropy similarity, stress mutation similarity, and fractal stability similarity). The process is stopped when the distance growth rate first drops below 10%. The autocorrelation coefficient of the feature matrix element sequence is calculated, and the time step at which it first drops to zero is selected as the delay parameter. The feature matrix is ​​arranged in chronological order. The phase space of each time point consists of the correlation strength values ​​within the current moment and the previous five delay periods. The diffusion rate of adjacent data points in the phase space is tracked. If adjacent points separate rapidly over time (for example, the separation amplitude per period exceeds twice the mean), it is marked as an anomaly. The separation amplitude of the last 200 periods is continuously recorded, and the average value plus three times the fluctuation range is used as the warning boundary.

[0080] The regional characteristic parameter group, temporal similarity parameter and time-varying threshold boundary are fused to form a creep feature vector set, specifically including: performing range normalization on the regional characteristic parameter group, assigning weights according to the dominance of energy transfer (wavelet energy entropy 0.4, stress mutation 0.3, fractal stability 0.3), and generating basic feature vectors; aligning the temporal similarity parameters through the dynamic time warping algorithm, calculating the weighted fusion value of the minimum cumulative distance and Pearson correlation coefficient between the three groups of regions, and dynamically adjusting the weights in combination with the silhouette coefficient generated by hierarchical clustering to form temporal correlation features; time-varying The threshold boundary captures abnormal marks (diffusion speed exceeds 2 times the mean and is marked as 1, otherwise 0) and warning boundary deviation (ratio of current value to mean) through phase space reconstruction to generate dynamic threshold features; finally, the basic feature vector, time series correlation feature and dynamic threshold feature are spliced ​​into a three-dimensional tensor according to the time step, and the principal component analysis is used to extract the first 60% of the variance principal component, and the time series dependency is integrated through the LSTM gating mechanism, in which the input gate retains the long-term trend of regional features, the forgetting gate attenuates the timeliness of temporal similarity, and the output gate combines the abnormal mark to generate a 12-dimensional creep feature vector (including 3 regions 4 parameter types abnormal state).

[0081] The monitoring method based on multi-dimensional feature fusion and dynamic threshold optimization has significantly improved the ability to identify early damage of the foundation pit support structure of the integrated pipeline corridor in soft soil areas. The traditional monitoring system relies on single-parameter static threshold analysis, which is easily interfered by noise and has difficulty in capturing low-amplitude, long-period creep characteristics. The present invention realizes the accurate characterization of progressive deformation through the joint analysis of wavelet energy entropy, stress mutation sensitivity factor and time-varying fractal stability parameter. The wavelet energy entropy can decompose strain signals of different frequencies and effectively distinguish noise from real creep response. The stress mutation sensitivity factor aligns the time series data through the dynamic time warping algorithm, which can identify local abnormal fluctuations in the stress distribution of the support structure. The time-varying fractal stability parameter is obtained from the variable The cumulative effect of creep is quantified from the perspective of self-similarity of shape patterns. The parameter similarities between regions are weighted and fused through a hierarchical clustering algorithm. The constructed 3×3 feature matrix comprehensively reflects the energy transfer correlation, stress synchronization characteristics and deformation synergy, overcoming the limitations of traditional methods of isolated parameter analysis. Compared with traditional static thresholds, the time-varying threshold boundary generated by phase space reconstruction can dynamically adjust the warning boundary by tracking abnormal fluctuations in the diffusion velocity of phase points to avoid false alarms or omissions caused by noise. When the separation amplitude of adjacent data points exceeds twice the historical mean, the system automatically marks the anomaly and combines the LSTM gating mechanism to fuse long-term trends and short-term mutation characteristics, significantly improving the sensitivity and timeliness of early damage identification.

[0082] Through the time series correlation analysis and creep mode collaborative early warning mechanism, the problem of multi-region linkage failure in the health monitoring of support structures in soft soil areas is effectively solved. The traditional system only focuses on the independent parameter changes of a single area, ignores the stress transfer and deformation synergy between regions, resulting in the global characteristics of early damage being masked by local noise. The present invention calculates the time series similarity parameters of the edge stress zone, the middle stable zone and the corner stress concentration zone, and quantifies the creep correlation strength of the three types of areas by weighted fusion. Among them, the dynamic time warping algorithm can align the time series data sequences of different regions and extract the fusion of the minimum cumulative distance and the linear correlation coefficient. The combined value accurately reflects the asynchronous characteristics of strain between regions. Combined with the weight distribution of the silhouette coefficient generated by hierarchical clustering, it further enhances the quantitative characterization ability of the coordination of deformation modes. In addition, the dual optimization of principal component analysis and LSTM network enables the 12-dimensional creep feature vector to simultaneously retain key information of long-term trends (such as creep accumulation rate) and short-term mutations (such as local stress concentration). At the same time, the input gate retains the long-term cumulative effect of wavelet energy entropy, the forgetting gate attenuates outdated time series similarity data, and the output gate generates a dynamic risk level based on the warning boundary deviation, realizing the transition from single parameter alarm to multi-mode collaborative warning.

[0083] In one case of this embodiment, the original data set is decomposed into three monitoring areas to obtain wavelet energy entropy parameters, stress mutation sensitivity factors, and time-varying fractal stability parameters to form a regional characteristic parameter group, including:

[0084] The original data set is subjected to noise separation to obtain the noise energy ratio parameter, specifically including: using multi-level wavelet decomposition technology, using db5 wavelet basis to decompose the original signal of the original data set into 5 layers, obtaining sub-band signals of different frequency bands, and identifying high-frequency noise components by calculating the sample entropy and variance contribution rate of the wavelet coefficients of each layer: when the sample entropy of a certain layer of sub-band is 0.5 and variance contribution rate When the variance contribution rate is 0.1, the layer is judged to be a high-frequency noise layer dominated by mechanical vibration. 0.3 and Pearson correlation coefficient When the value is 0.2, it is determined to be a low-frequency interference layer dominated by environmental noise. Soft threshold processing is performed on the mechanical noise layer, and the threshold is 0.6 times the absolute value of the median coefficient of this layer. The mechanical impact component is eliminated by wavelet reconstruction. The improved spectral subtraction method is used for the environmental noise layer. The noise power spectrum is estimated based on the Wiener filter. Energy compensation is performed in the time-frequency domain to retain the effective signal characteristics. The energy entropy of the sub-band signal at each processing stage is calculated, and the noise energy ratio parameter is defined as: mechanical noise energy ratio (Mechanical layer sub-band energy × entropy weight) Total energy, ambient noise ratio (Environmental layer sub-band energy Entropy Weight) Total energy, where the entropy weight is dynamically adjusted according to the entropy of the subband samples. The weight increases by 5% for every 0.1 increase in entropy value. The noise energy ratio parameter in the range of 0-1 is then obtained through energy normalization.

[0085] The original data set is decomposed according to the noise energy ratio parameter to obtain the energy weight parameter of each frequency band, forming a weighted initial feature set. Specifically, the original signal is divided into continuous frequency bands using a bandpass filter bank. Each frequency band corresponds to a specific frequency range. Based on the distribution characteristics of mechanical noise and environmental noise in the noise energy ratio parameter, noise suppression processing is performed on high-frequency bands (above 500Hz). When the mechanical noise energy ratio of a frequency band is detected to exceed a preset threshold, the energy weight of the frequency band is adjusted using a dynamic attenuation factor. The attenuation strength is positively correlated with the mechanical noise ratio. At the same time, the frequency band energy of the effective vibration component is retained. Environmental noise compensation is performed on low-frequency bands (below 200Hz). The compensation coefficient is dynamically adjusted based on the calculated frequency band signal-to-noise ratio. When the signal-to-noise ratio is lower than the preset threshold, the environmental noise power spectrum is estimated using a Wiener filter. Energy compensation is performed in the time-frequency domain to improve the signal-to-noise ratio of the effective signal. The effective energy weight of each frequency band is calculated using an energy redistribution algorithm. The effective energy of all frequency bands is then normalized to form a weighted initial feature set reflecting the characteristic strength of each frequency band.

[0086] The weighted initial feature set is calculated to obtain the wavelet energy entropy parameter of the edge stress area, specifically including: the energy value of the wavelet transform of the uth layer (u ranges from 1 to 3) The fractal box dimension of the wavelet coefficients of this layer Multiply them together to get the energy-fractal dimension product term , calculate the Morlet wavelet basis function In the scale factor interval Points on , get the integral result, and multiply the energy-fractal dimension product Divide the integral result to get the ratio of energy-fractal to wavelet basis integral , the ratio of energy-fractal to wavelet basis integral is taken as To the power of energy-fractal entropy , sum the three-layer results of u from 1 to 3, and get the three-layer energy-fractal entropy weight sum , multiply the three-layer energy-fractal entropy weight by , and obtain the wavelet energy entropy parameter of the edge stress area , when applied specifically, it can be achieved through the following calculation formula, for example:

[0087] ;

[0088] Where, represents the wavelet energy entropy parameter, represents the energy value of the wavelet transform of the u-th layer, represents the fractal box dimension of the u-th layer wavelet coefficients, represents the maximum scale factor, represents the minimum scale factor, represents the Morlet wavelet basis function, represents the differential symbol of the scale factor a in the continuous wavelet transform, represents the dynamic attenuation factor;

[0089] The weighted initial feature set is decomposed and adjusted to obtain the stress mutation sensitivity factor in the central stable area, specifically including: The actual stress change The difference between the two values ​​is divided by 0.5 times the difference to obtain the ratio term. , square the comparison term and multiply it by And take the exponent to get the exponential term , the fractal dimension of the stress series Divide by 2 and add to 1 to get the fractal dimension correction term , multiply the exponential term by the fractal dimension correction term to obtain the stress mutation sensitivity factor , when applied specifically, it can be achieved through the following calculation formula, for example:

[0090] ;

[0091] Where, represents the stress mutation sensitive factor, Indicates the LSTM predicted stress change, represents the actual stress change, represents the fractal dimension of stress series, exp represents the exponential function;

[0092] The fractal dimension of the weighted initial feature set is calculated to obtain the time-varying fractal stability parameters of the corner stress concentration area, which specifically includes: the Hurst exponent of the kth window is converted to Multiply , and get the numerator , and Add together to get the denominator , divide the numerator by the denominator to get the proportional term , multiply the proportional term by the Gamma function , we get the contribution of the kth window to the stability of the time-varying fractal , sum all the contribution terms from 1 to the total number of time window partitions K, and get the time-varying fractal stability parameter , when applied specifically, it can be achieved through the following calculation formula, for example:

[0093] ;

[0094] Where, represents the time-varying fractal stability parameter, represents the Hurst exponent of the kth window, represents the coefficient of variation of deformation rate within the window, e represents the base of natural logarithm, represents the Gamma function, Indicates the total number of time window divisions, represents the time-related parameters of the k-th time window;

[0095] After normalization, the wavelet energy entropy parameters, stress mutation sensitivity factors and time-varying fractal stability parameters are weightedly fused to generate a regional characteristic parameter group, including: performing range normalization on the wavelet energy entropy parameters, stress mutation sensitivity factors and time-varying fractal stability parameters, and using the noise energy proportion parameters of each region as the dynamic weight coefficient to establish a weighted fusion rule, specifically: the weight of the edge stress zone is dominated by the mechanical noise energy proportion (proportion When the weight is 60%, it is 0.5, accounting for At 60%, the weight value decreases by 0.05 for every 10% decrease in the proportion of mechanical noise, maintaining a minimum baseline weight of 0.2). The central stable zone adopts a weighted average of the stress mutation sensitivity factor and fractal stability (weight ratio 3:2). The corner stress concentration zone dynamically adjusts the weight according to the coefficient of variation of the time-varying fractal stability parameter (the weight increases by 15% for every 0.1 increase in the coefficient). Through the feature splicing algorithm, the normalized wavelet energy entropy parameter, stress mutation sensitivity factor, and time-varying fractal stability parameter are aligned according to the time step. Principal component analysis is used to extract the principal components contributing to the first 60% of the variance. The temporal dependencies are then integrated through the LSTM gating mechanism to generate a 12-dimensional regional feature parameter group containing spatiotemporal feature associations.

[0096] Through multi-level noise separation and dynamic feature fusion, the sensitivity of identifying early damage to support structures has been significantly improved, which is specifically manifested in three aspects: the dual-criteria noise stratification technology based on sample entropy and variance contribution rate, combined with improved spectral subtraction and adaptive threshold processing, achieves accurate separation of mechanical vibration and environmental interference, and effectively suppresses the masking effect of noise on weak deformation characteristics. Secondly, the energy-fractal entropy weight parameter and time-varying fractal stability parameter are used to construct a multi-scale feature expression method. Through the dynamic coupling of fractal dimension and Hurst exponent, the system's sensitivity to creep rate detection is improved, and low-amplitude and long-period deformation characteristics that are difficult to identify with traditional methods are successfully captured. Finally, based on the dynamic weight allocation mechanism of noise proportion and the LSTM-PCA fusion algorithm, the spatiotemporal correlation analysis of the characteristics of the three monitoring areas is realized. Compared with the traditional static threshold method, the false alarm rate is reduced, the warning response time is improved, and it is convenient for early diagnosis of progressive damage to support structures in soft soil areas.

[0097] In one case of this embodiment, the creep feature vector set is analyzed to obtain a monitoring regionalized time-varying threshold set, including:

[0098] The risk coefficients of each monitoring area are set, with the corner stress concentration area being 0.6, the edge stress area being 0.3, and the central stable area being 0.1. Specifically, they include:

[0099] In soft soil environments, due to the low shear strength of the soil (usually 5-15 kPa), the stress concentration coefficient of the geometric mutation area in the corner stress concentration zone can reach 2-3, which is much higher than that in other areas. The stress in the corner area continues to increase over time, and the creep characteristics of the soft soil exacerbate local deformation. Failure of the corner stress concentration area may cause cascading instability of the support system (such as overall collapse due to broken corner braces). Therefore, a high weight (0.6) is required to ensure that anomalies (such as stress mutations >15%) are quickly identified and trigger active intervention.

[0100] In a soft soil environment, the edge stress zone directly bears the lateral pressure of the soil. However, the rheological properties of soft soil make the stress distribution relatively uniform and the failure rate is slow. The failure of the edge zone is often manifested as tilting of support piles or cracking of crown beams. The risk consequences are controllable. Usually, attention should be paid to the effects of long-term creep accumulation. Therefore, a medium weight (0.3) is set.

[0101] In a soft soil environment, the central stable zone is subjected to the soil's own weight and uniformly distributed load. Compared with the stress concentration zone at the corner and the edge stress zone, the probability of local failure is extremely low, so a low weight (0.1) is set.

[0102] Based on the risk coefficient of each monitoring area, the creep eigenvector of the monitoring area is weighted and fused to generate a regional coupling weight matrix reflecting the coupling effect of the characteristics of different monitoring areas. Specifically, the risk coefficient is used as the dynamic weight coefficient, the weight of the corner stress concentration area is 0.6, the edge stress area is 0.3, and the central stable area is 0.1. The eigenvector of each area is multiplied element by element with the corresponding risk coefficient through the feature-level linear weighting algorithm, where each parameter value of the eigenvector of the corner area is multiplied by 0.6, the parameter value of the edge area is multiplied by 0.3, and the parameter value of the central area is multiplied by 0.1. The three weighted eigenvectors are time series aligned, and the principal component contributing to the first 60% of the variance is extracted by principal component analysis. The orthogonal feature set is generated by the maximum variance rotation method. The principal component features of the three areas are superimposed according to the risk weight. The principal component of the corner area is multiplied by 0.6 and placed in the first row of the matrix, the principal component of the edge area is multiplied by 0.3 and placed in the second row, and the principal component of the central area is multiplied by 0.1 and placed in the last row, forming a regional coupling weight matrix with 3 rows and 12 columns.

[0103] Based on the regional coupling weight matrix, the creep eigenvector set is statistically analyzed to generate a dynamic threshold benchmark curve.

[0104] The dynamic threshold baseline curve is corrected according to the time series similarity parameters to obtain the set of time-varying thresholds for monitoring regions. Specifically, based on the time series similarity parameters of each regional pair in the feature matrix (wavelet energy entropy similarity, stress mutation similarity, fractal stability similarity), the time axis of the baseline curve is aligned using a dynamic time warping algorithm to eliminate the time offset effect of monitoring data in different regions. By calculating the Euclidean distance between the similarity parameter of each time step and the corresponding moment of the baseline curve, a distance matrix is ​​constructed, in which the distance value reflects the degree of deviation between the baseline curve and the actual characteristics of the region. The silhouette coefficient generated by hierarchical clustering is then used as the weight coefficient to dynamically weight the abnormal points in the distance matrix. For clusters with silhouette coefficients higher than the threshold, the weight distribution of the corresponding region is reduced (the reduction is 1.5 times the silhouette coefficient value), otherwise the weight is increased. The modified baseline curve is fitted by the weighted least squares method to make the curve shape more consistent with the characteristic distribution of high-similarity regions. The Pearson correlation coefficient in the time series similarity parameter is used as a dynamic adjustment factor to implement slope compensation on the modified baseline curve. When the correlation coefficient of the region pair exceeds 0.7, the rising rate of the baseline curve is increased by 0.3 of the correlation coefficient value. When the correlation coefficient is lower than 0.4, the fluctuation amplitude of the baseline curve is reduced by 0.15. Finally, a time-varying threshold set containing 9 region nodes is formed.

[0105] Through regionalized dynamic threshold analysis of creep eigenvectors, the ability to identify early damage to the foundation pit support structure of integrated pipeline corridors in soft soil areas has been significantly improved. Based on the risk grading mechanism of the monitoring area (0.6 for corner stress concentration areas, 0.3 for edge stress areas, and 0.1 for central stable areas), and combined with the engineering characteristics of soft soil's low shear strength and stress concentration factor, targeted monitoring enhancement of high-risk areas is achieved. A regional coupling weight matrix is ​​generated through a weighted fusion algorithm, and principal component analysis is performed to extract the high-frequency stress mutations (>15%) in the corner area and the long-term creep cumulative effect. This effectively suppresses the interference of background noise on low-amplitude creep signals, improves the detection sensitivity of small strains, and solves the mismatch between long-term creep deformation and static thresholds in soft soil. At the same time, hierarchical clustering and dynamic weighting of the contour coefficient can adaptively adjust the shape of the threshold curve to reduce the false alarm rate caused by soil heterogeneity.

[0106] Through multi-dimensional feature fusion and dynamic threshold optimization, an active intervention method for support structure health monitoring in soft soil environments was constructed. Based on the principal component superposition mechanism of the regional coupling weight matrix and the spatiotemporal correlation between soft soil creep parameters and stress distribution, the precise positioning of progressive damage such as support pile tilt and crown beam cracking was achieved. The weighted least squares fitting and slope compensation algorithms enabled the time-varying threshold set to fit the nonlinear characteristics of soft soil creep. For example, when stress suddenly changes in the corner area, the rising rate of the baseline curve is automatically increased (by 30% when the correlation coefficient is >0.7), ensuring a rapid response to high-risk events. By dynamically adjusting the support axial force monitoring threshold, the deep horizontal displacement caused by local creep accumulation is avoided, meeting the real-time requirements of long-term monitoring in soft soil areas.

[0107] In one case of this embodiment, based on the regional coupling weight matrix, statistics are collected on the creep eigenvector set to generate a dynamic threshold reference curve, including:

[0108] The creep eigenvector set is calculated using an adaptive sliding window algorithm to generate a two-parameter eigenvector. Specifically, the adaptive sliding window algorithm is used to dynamically determine the initial window length based on the information entropy of the creep eigenvector sequence. Specifically, the Shannon entropy of each dimension is calculated for the eigenvector of each time step, and the position where the absolute value of the second-order difference of the entropy value sequence exceeds the preset threshold (0.15) is used as the trigger point for adjusting the window boundary. When the window slides along the time axis, the dynamic standard deviation of the eigenvector in the current window is calculated dimension by dimension using the rolling standard deviation algorithm. That is, for the numerical sequence of each dimension, the standard deviation of the latest data point is updated in real time to eliminate the influence of the data points that have moved out of the window. The eigenvector sequence of each dimension in the window is linearly fitted using the least squares method. The trend slope is obtained by solving the straight line parameter with the minimum sum of squared errors. The dynamic standard deviation corresponding to each dimension is combined with the trend slope to form a two-parameter eigenvector reflecting the data fluctuation characteristics and change trend.

[0109] The dual-parameter eigenvectors were analyzed based on the regional coupling weight matrix to obtain the soft soil creep stage identification factor. Specifically, based on the regional coupling weight matrix, the dual-parameter eigenvectors were weighted dimension by dimension according to the regional risk weight: the characteristic dimension corresponding to the corner stress concentration area was multiplied by a weight of 0.6, the dimension of the edge stress area was multiplied by a weight of 0.3, and the dimension of the central stable area was multiplied by a weight of 0.1, highlighting the impact of the fluctuation and trend characteristics of the high-risk area on the results. After weighting, principal component analysis was performed on the 12-dimensional eigenvectors. By calculating the eigenvalues ​​of the covariance matrix, the principal components with a cumulative variance contribution rate of 80% were extracted to construct a low-dimensional space containing the main information of the creep characteristics. In the principal component space, K-means++ was used. The algorithm clusters the data points, with a preset number of clusters K = 3 (corresponding to the initial stage, stable stage, and accelerated stage of soft soil creep). The silhouette coefficient of each data point is calculated (a measure of the closeness of the sample to its cluster and the degree of separation from other clusters). When the average silhouette coefficient is greater than 0.5, the clustering is considered valid. Otherwise, the initial center is adjusted and the cluster is re-clustered. Based on the fluctuation-trend characteristics of each cluster data (for example, the acceleration stage usually has a high trend slope and dynamic standard deviation), the cluster labels are manually assigned, namely initial creep, stable creep, and accelerated creep. The Euclidean distance of the principal component feature point at the current moment to the three cluster centers is calculated and normalized by the inverse of the distance to generate a soft soil creep stage identification factor containing three elements.

[0110] The dual-parameter eigenvectors are mapped according to the soft soil creep stage identification factors to generate a dynamic threshold benchmark curve, specifically including: classifying the dual-parameter eigenvectors (dynamic standard deviation, trend slope) by stage according to the soft soil creep stage identification factors: extracting the data of each stage, calculating the dual-parameter mean and standard deviation of the initial stage, stable stage, and accelerated stage respectively, and according to the identification factors at the current moment, weighting the fusion of the benchmark templates of each stage according to the membership ratio (the initial stage mean and standard deviation are respectively calculated). 0.1 Stable phase mean 0.6 Acceleration phase mean 0.3), get the mean of the current dynamic threshold, fuse the standard deviations of each stage, get the range of fluctuation, and use the exponentially weighted moving average algorithm to smooth the fused mean and standard deviation series to suppress random noise and generate a data set containing the mean center line and Dynamic threshold benchmark curve for the 2 standard deviation fluctuation band.

[0111] An adaptive sliding window algorithm dynamically determines the window length, captures data fluctuation characteristics and changing trends in real time, and generates a dual-parameter eigenvector. This avoids the missed detection of low-amplitude, long-period deformations by a fixed window. The dual-parameter eigenvector is weighted with the help of a regional coupling weight matrix to highlight the influence of high-risk regional characteristics. Combining principal component analysis with K-means++ clustering, this method accurately identifies the creep stage of soft soil and effectively distinguishes between the initial, stable, and accelerated stages. This overcomes the limitations of independent analysis of a single parameter. The generated dynamic threshold baseline curve integrates the mean and standard deviation of the characteristics of each stage and suppresses noise through time series smoothing to form a dynamic threshold containing the mean centerline and fluctuation band. This overcomes the problem of traditional static thresholds that over-rely on fixed values ​​and are easily masked by background noise. It can promptly detect subtle changes in early damage to the support structure and improve the health monitoring system's ability to capture progressive deformation. This provides more reliable technical support for the safety assessment and early warning of foundation pit support structures in soft soil areas, ensuring the long-term stability and safety of integrated pipeline corridor projects.

[0112] In one case of this embodiment, progressive damage determination is performed on the creep feature vector set and the monitoring regionalized time-varying threshold set, and the damage position of the monitoring target object is located and an early warning is issued, including:

[0113] Based on the risk coefficient, the creep characteristic vector set and the monitoring regional time-varying threshold set are regionally operated to generate the regional damage deviation. Specifically, based on the risk coefficient, the Euclidean distance algorithm is used to calculate the original deviation between the creep characteristic vector of each monitoring area and the corresponding time-varying threshold. The original deviation is weighted and fused with the risk coefficient of 0.6 for the corner stress concentration area, 0.3 for the edge stress area, and 0.1 for the central stable area. The weighted deviation is multiplied element by element through the feature-level linear weighting algorithm, and then the weighted deviation is normalized to generate the regional damage deviation reflecting the damage degree of each area.

[0114] The regional damage deviation is corrected based on the time series similarity parameter to obtain the cross-regional damage correlation value. Specifically, based on the time series similarity parameter, the wavelet energy entropy, stress mutation, and fractal stability similarity parameters between the three groups of regions are normalized. The comprehensive similarity of the regional pairs is calculated using preset weights (energy entropy 0.4, stress mutation 0.3, fractal stability 0.3). The regional damage deviation time axis is aligned using the dynamic time warping algorithm. The deviation of the regional pairs is weighted by a weighted product operation using the comprehensive similarity. After normalization, a cross-regional damage correlation value reflecting the degree of damage correlation between the regions is generated.

[0115] The regional damage deviation and cross-regional damage correlation values ​​are weighted and fused to generate a comprehensive damage location index. Specifically, the regional damage deviation and cross-regional damage correlation values ​​are subjected to range normalization processing respectively, and the risk coefficient of each monitoring area (0.6 for the corner area, 0.3 for the edge area, and 0.1 for the central area) is used as the weight of the regional damage deviation. The single-region deviation is weighted using a feature-level linear weighting algorithm. For the cross-regional damage correlation value, the preset weights of the time series similarity parameters (energy entropy 0.4, stress mutation 0.3, and fractal stability 0.3) are used to weight the correlation values ​​of the three groups of regions. The two types of weighted results are fused element by element through matrix dot multiplication. After normalization, the index is based on the regional damage dominance (weight 0.7). The proportion of cross-region correlation correction weights (weight 03) is linearly superimposed to generate a comprehensive damage location index in the range of 0-1;

[0116] When the comprehensive damage location index exceeds the regional time-varying threshold of the corresponding monitoring area, a graded warning is triggered for the monitoring area, and the damage location is located by the index peak.

[0117] A regional damage deviation calculation method based on risk coefficient weighting, combined with differentiated risk weights for corner stress concentration zones, edge stress zones, and central stable zones, achieves precise matching of creep eigenvectors with dynamic thresholds. This solves the problem of low-amplitude deformations being masked by noise and significantly improves the ability to identify early micro-strain accumulation. A temporal similarity parameter is used to perform cross-regional correlation correction on the regional damage deviation. By integrating multi-dimensional similarity parameters such as wavelet energy entropy, stress mutation, and fractal stability, and combining them with a dynamic time warping algorithm to achieve time axis alignment, this method captures the spatiotemporal correlation of damage evolution between regions, avoiding the one-sidedness of single-parameter independent analysis. This upgrades damage assessment from isolated threshold comparison to multi-parameter coupled analysis. By organically combining regional damage dominance with cross-regional correlation correction, a quantitative assessment method in the 0-1 range is formed, which not only highlights the damage sensitivity of key areas but also reflects the coordinated evolutionary characteristics of overall structural damage. When the index exceeds the regionalized time-varying threshold, graded warnings and precise damage location can be achieved, effectively improving the timeliness and accuracy of health monitoring of support structures in soft soil areas.

[0118] In one aspect of this embodiment, locating the damage position by using the index peak value includes:

[0119] The comprehensive damage location index is spatially discretized to generate the local peak coefficient of each monitoring area. Specifically, when the comprehensive damage location index is spatially discretized, the spatial unit is first divided based on the grid monitoring area, and the sliding window neighborhood analysis method is used to traverse each area. For each monitored spatial unit, its neighborhood window is defined as 3 3 grids, using the local maximum detection algorithm to extract the highest index value in the window as a candidate peak point, and calculate the relative intensity difference ratio of the index of the adjacent 8 spatial units in its neighborhood with the candidate point as the center. Combined with the Gaussian weighted kernel function (standard deviation of 0.5), spatial smoothing is performed to generate the initial peak coefficient. The initial coefficient of each area is mapped to the range of 0-1 through range normalization. After normalization, the local peak coefficient of each monitoring area is obtained;

[0120] The local peak coefficient is calculated according to the risk coefficient of each monitoring area to obtain the weighted peak response value. Specifically, the following steps are taken: the local peak coefficient of the corner stress concentration area is multiplied by 0.6, the coefficient of the edge stress area is multiplied by 0.3, and the coefficient of the central stable area is multiplied by 0.1 using the characteristic-level linear weighting algorithm to form the risk-weighted initial response vector. The initial response vector is tested for spatial consistency using the sliding window neighborhood analysis method. The spatial correlation weight of the inter-regional response value is calculated using the Gaussian kernel function (with a standard deviation of 0.5) for adjacent cells in a 3×3 grid. The initial response vector is multiplied element-by-element by the spatial weight matrix in combination with the range normalization algorithm. The characteristic components contributing to the first 60% of the variance are extracted using principal component analysis. Finally, the components are superimposed and normalized to the range of 0-1 to obtain the weighted peak response value reflecting the association between risk weight and space.

[0121] The extreme point position of the weighted peak response value is analyzed to generate the location and confidence parameters of the damage position, which includes: performing spatial discretization grid processing on the weighted peak response value, dividing the monitoring area into regular grid units and assigning row and column indexes ( ), through sliding window neighborhood analysis, in each 3 3. The local maximum detection algorithm is used in the grid window to identify the unit with the highest response value as the extreme point, and its row and column index ( ) is determined as the coordinate of the damage location. For the extreme point, the spatial weights of its 8 neighborhood units are calculated: with the extreme point as the center, the neighborhood weights are generated according to the distance attenuation rule based on the Gaussian kernel function (standard deviation is 0.5). The closer the distance, the higher the weight. The extreme point response value is weighted in combination with the regional risk coefficient (0.6 in the corner area, 0.3 in the edge area, and 0.1 in the middle area) to highlight the extreme value intensity in the high-risk area. The weighted response value is then normalized for the range and mapped to the 0-1 interval. By comparing the degree of difference between the extreme point response value and the neighborhood mean, the difference ratio is used as the basis, multiplied by the spatial weight and the regional risk weight, and finally a positioning confidence parameter in the 0-1 interval is generated. The higher the value, the stronger the positioning reliability.

[0122] Through gridded spatial discretization and 3×3 sliding window analysis, combined with the spatial smoothing technology of Gaussian weighted kernel function, the interference of background noise on low-amplitude signals is effectively filtered out, so that the local peak coefficient calculation can not only capture the abnormal characteristics of small strain accumulation, but also suppress the false triggering of random noise, solving the problem that a single parameter threshold in traditional methods is difficult to distinguish between noise and real damage signals. Through the dual weighting mechanism of regional risk coefficient and spatial correlation weight, the damage sensitivity of high-risk areas such as corner stress concentration areas is highlighted in the peak response calculation, and the spatial correlation of adjacent units is quantified by Gaussian kernel function to avoid the abnormality caused by single regional outliers. Misjudgment, the damage coordinates are located through the extreme point row and column index, and the location confidence parameters in the range of 0-1 are generated by combining the regional risk weight and the Gaussian distance decay rule, forming a three-level judgment system of spatial positioning-risk weighting-confidence assessment. It not only realizes fine positioning, but also upgrades the discrete point alarm of traditional monitoring to regional damage positioning with spatial correlation through the dual constraints of risk weight and spatial correlation, effectively solving the problem of low-rate and long-period deformation monitoring caused by soft soil creep, providing reliable technical support for the accurate diagnosis and safety warning of early damage of support structures, and significantly improving the intelligent level of engineering structure health monitoring.

[0123] In one case of this embodiment, when the comprehensive damage localization index exceeds the regional time-varying threshold of the corresponding monitoring area, a graded warning for the monitoring area is triggered, including:

[0124] The comprehensive damage location index in the corner stress concentration area 1.0 and continuing If the time is more than one hour, a level one warning will be triggered;

[0125] Comprehensive damage location index in the edge stress zone 1.1 and continuing If the time is more than one hour, a second-level warning will be triggered;

[0126] Comprehensive damage localization index in the central stable zone 1.2 and continue Hours or more, triggering the third-level warning;

[0127] Among them, the severity of warning is level one warning > level two warning > level three warning;

[0128] If the cross-regional damage correlation values ​​of adjacent monitoring areas are all greater than 0.7, a coordinated warning is triggered;

[0129] In one case of this embodiment, if the cross-region damage correlation values ​​of adjacent monitoring areas are all greater than 0.7, a coordinated warning is triggered, including:

[0130] If the cross-region damage correlation values ​​of any two adjacent monitoring areas in the edge stress zone, corner stress concentration zone and central stable zone are all greater than 0.7, the coordinated first-level warning is triggered. The adjacent monitoring areas in the foundation pit structure are only the corner stress concentration zone and the edge stress zone, and the corner stress concentration zone and the central stable zone;

[0131] If the cross-regional damage correlation values ​​of the edge stress zone, corner stress concentration zone, and central stable zone are all greater than 0.7, a coordinated secondary warning is triggered;

[0132] Among them, the severity of the warning is coordinated level 2 warning > coordinated level 1 warning.

[0133] By constructing a hierarchical warning and coordinated enhancement mechanism, the problem of delayed response of traditional static threshold monitoring to early damage of soft soil creep is accurately solved. According to the differentiated risk levels of corner stress concentration areas, edge stress areas, and central stable areas, stepped warning thresholds (1.0, 1.1 and 1.2) and durations (1, 2 and 4 hours) are set to achieve more sensitive early warning responses in high-risk areas, avoid false triggering in low-risk areas, and significantly improve the targeted nature of progressive deformation monitoring.

[0134] Through the collaborative warning mechanism triggered by cross-region damage correlation values, a single-region hierarchical warning system is established by identifying the damage correlation between corner-edge and corner-center regions. The dual judgment system with cross-regional collaborative enhancement triggers a collaborative first-level warning when the correlation values ​​of any two adjacent areas exceed the standard, and upgrades to a collaborative second-level warning when all three areas are correlated. It effectively captures the spatial evolution characteristics of structural damage and solves the one-sided problem of traditional monitoring and isolated analysis of single parameters. It not only highlights the immediate warning sensitivity of high-risk areas, but also quantifies the collaborative development trend of damage through cross-regional correlation measurement, transforming passive threshold alarms into intelligent warnings that combine risk levels and spatial correlations, thereby improving the timeliness and reliability of health monitoring of support structures in soft soil areas.

[0135] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A health monitoring system for foundation pit support structures of a comprehensive pipe gallery, characterized in that: include: The monitoring unit is used to divide the monitoring target object into three monitoring areas, collect heterogeneous monitoring data from each monitoring area in real time, and generate a pre-processed raw data set. The monitoring target object is the support structure of the foundation pit in the soft soil area. The three monitoring areas include: Edge stress zone, middle stable zone and corner stress concentration zone; Among them, the edge stress zone is the edge where the support structure contacts the surrounding soil; The middle stable zone is the middle part of the support structure; The corner stress concentration area is the corner of the foundation pit plane; The mining unit is used to perform feature mining on the original data set and generate a creep feature vector set, including: The original data set is decomposed into three monitoring areas to obtain wavelet energy entropy parameters, stress mutation sensitivity factors and time-varying fractal stability parameters, forming a regional characteristic parameter group, including: Perform noise separation on the original data set to obtain the noise energy ratio parameter; Decompose the original data set according to the noise energy ratio parameter to obtain the energy weight parameter of each frequency band and form the weighted initial feature set; The weighted initial feature set is calculated to obtain the wavelet energy entropy parameters of the edge stress area; The weighted initial feature set is decomposed and adjusted to obtain the stress mutation sensitivity factor in the central stable zone; The fractal dimension of the weighted initial feature set is calculated to obtain the time-varying fractal stability parameters of the corner stress concentration zone. The wavelet energy entropy parameter, stress mutation sensitivity factor and time-varying fractal stability parameter are normalized and then weighted fused to generate a regional characteristic parameter group. Based on the regional characteristic parameter group, the temporal similarity parameters between each monitoring area are calculated; Decompose the time series similarity parameters and construct a feature matrix reflecting the time series correlation of the monitoring area; Reconstruct the phase space of the feature matrix to generate a time-varying threshold boundary; The regional feature parameter group, the temporal similarity parameter and the time-varying threshold boundary are fused to form a creep feature vector set; An analysis unit, used for analyzing the creep characteristic vector set to obtain a monitoring regionalized time-varying threshold set; The judgment unit is used to perform progressive damage judgment on the creep feature vector set and the monitoring regional time-varying threshold set, locate the damage position of the monitoring target object and issue an early warning, including: Based on the risk coefficient, the creep characteristic vector set and the monitoring regional time-varying threshold set are regionally operated to generate the regional damage deviation. The regional damage deviation is corrected according to the time series similarity parameter to obtain the cross-regional damage correlation value; The regional damage deviation and cross-regional damage correlation values ​​are weighted and integrated to generate a comprehensive damage location index; When the comprehensive damage location index exceeds the regional time-varying threshold of the corresponding monitoring area, a graded warning is triggered for the monitoring area, and the damage location is located by the index peak.

2. A health monitoring system for foundation pit support structures of a comprehensive pipe gallery according to claim 1, characterized in that: By analyzing the creep characteristic vector set, a set of time-varying thresholds for monitoring regions is obtained, including: Set the risk factor for each monitoring area; Based on the risk coefficient of each monitoring area, the creep characteristic vector of the monitoring area is weighted and fused to generate a regional coupling weight matrix that reflects the coupling effect of the characteristics of different monitoring areas. Based on the regional coupling weight matrix, the creep eigenvector set is statistically analyzed to generate a dynamic threshold benchmark curve. The dynamic threshold benchmark curve is modified according to the time series similarity parameter to obtain the monitoring regional time-varying threshold set.

3. The health monitoring system for the foundation pit support structure of a comprehensive pipe gallery according to claim 1 is characterized in that: Based on the regional coupling weight matrix, the creep eigenvector set is statistically analyzed to generate a dynamic threshold benchmark curve, including: The creep eigenvector set is calculated by an adaptive sliding window algorithm to generate a two-parameter eigenvector; The double-parameter eigenvector is analyzed based on the regional coupling weight matrix to obtain the soft soil creep stage identification factor. The dual-parameter eigenvector is mapped according to the soft soil creep stage identification factor to generate a dynamic threshold benchmark curve.

4. The health monitoring system for the foundation pit support structure of a comprehensive pipe gallery according to claim 1 is characterized in that: Locating damage locations by index peaks includes: Perform spatial discretization on the comprehensive damage location index to generate the local peak coefficient of each monitoring area; The local peak coefficient is calculated according to the risk factor of each monitoring area to obtain the weighted peak response value; The extreme point position of the weighted peak response value is analyzed to generate the location and confidence parameters of the damage position.

5. The health monitoring system for foundation pit support structure of a comprehensive pipe gallery according to claim 1 is characterized in that: When the comprehensive damage location index exceeds the regional time-varying threshold of the corresponding monitoring area, a graded warning is triggered for the monitoring area, including: When the comprehensive damage location index in the corner stress concentration area is greater than 1.0 and lasts for more than 1 hour, a level 1 warning is triggered; When the comprehensive damage location index in the edge stress zone is greater than 1.1 and lasts for more than 2 hours, a second-level warning is triggered; In the central stable zone, when the comprehensive damage localization index is greater than 1.2 and lasts for more than 4 hours, a level 3 warning is triggered; Among them, the severity of warning is level one warning > level two warning > level three warning; If the cross-regional damage correlation values ​​of adjacent monitoring areas are all greater than 0.7, a coordinated warning will be triggered.

6. The health monitoring system for the foundation pit support structure of the integrated pipe gallery according to claim 5 is characterized in that: If the cross-regional damage correlation values ​​of adjacent monitoring areas are all greater than 0.7, a coordinated warning is triggered, including: If the cross-regional damage correlation values ​​of any two adjacent monitoring areas in the edge stress zone, corner stress concentration zone and central stable zone are all greater than 0.7, a coordinated level 1 warning will be triggered; If the cross-regional damage correlation values ​​of the edge stress zone, corner stress concentration zone, and central stable zone are all greater than 0.7, a coordinated secondary warning is triggered; Among them, the severity of the warning is coordinated level 2 warning > coordinated level 1 warning.

Images