Trend analysis method for long-term monitoring data of lightweight steel arch

Abstract

This invention discloses a trend analysis method for long-term monitoring data of lightweight steel arches, relating to the field of structural health monitoring technology. The method includes the following steps: acquiring the steel arch strain sequence and arch position mileage data of the target object; performing arch position mapping segmentation on the steel arch strain sequence based on the arch position mileage data to generate arch-specific strain segments; and performing symmetrical convergence on the arch-specific strain segments to generate a trend matrix. This scheme generates a side weight matrix by performing dual-gated side weight reconstruction on the arch position mileage data and the trend matrix, and generates a common-mode trend sequence through neighborhood consistency convergence. This achieves weakening of distant accidental similarity and suppression of near-distance morphological differences, allowing regional common convergence to be centrally expressed. The common-mode trend sequence and the trend matrix are differentially separated to generate a differential-mode trend sequence, and a consensus inflection point set is generated through graph inflection point consensus screening. This achieves the removal of regional covariance and the filtering of false inflection points through neighborhood mutual verification.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring technology, specifically a method for trend analysis of long-term monitoring data of lightweight steel arches. Background Technology

[0002] With the expansion of underground salt mine resource development, salt mine transport roadways, as key channels in the mining production system for transportation, ventilation, and personnel passage, are prone to long-term creep of the surrounding rock, hydrothermal coupling, and brine erosion, which can lead to the accumulation of roadway deformation and the evolution of the stress on the support structure. Under this technical environment, lightweight steel arches have the advantage of reducing construction and maintenance costs due to the optimization of materials and structural design. However, their monitoring system is often affected by factors such as the limited number of sensor points, data noise, and communication limitations, which makes the trend analysis of long-term monitoring data face higher requirements for reliability and interpretability.

[0003] In existing technologies, trend analysis of long-term monitoring data for steel arches typically involves introducing monitoring information from adjacent arches on top of single-arch processing. The trend results of adjacent arches are smoothed or validated for consistency according to fixed spatial adjacency relationships to improve the ability to identify local anomalies. This approach can suppress single-point noise and enhance the spatial continuity of anomaly location to a certain extent when the arch group is densely distributed and the adjacency relationship is stable over a long period. However, in underground salt mine environments, the trends of adjacent arches diverge at different times due to surrounding rock conditions, construction differences, and changes in force transmission. Fixed aggregation forcibly mixes dissimilar neighborhoods, easily smoothing out real anomalies and generating false consistency. Consequently, inflection point detection based on aggregation results becomes more dependent on the amplitude of single-sequence mutations. Under the influence of salt spray and dust peaks, noise, and changes in sampling intervals, false inflection points are prone to occur, and there is a lack of consensus verification across the arch group. This leads to unstable anomaly triggering, making it difficult to characterize the propagation pattern of anomalies within the arch group, ultimately resulting in inaccurate trend analysis results for lightweight steel arch groups. Summary of the Invention

[0004] The purpose of this invention is to provide a method for trend analysis of long-term monitoring data of lightweight steel arches, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, this invention discloses a trend analysis method for long-term monitoring data of lightweight steel arches, applied to the trend analysis of continuously deployed lightweight steel arch groups in underground salt mine transportation tunnels, comprising the following steps:

[0007] Obtain the strain sequence and arch position mileage data of the target object's steel arch;

[0008] Based on the arch position mileage data, the steel arch strain sequence is subjected to arch position mapping and segmentation processing to generate arch strain segments, and symmetrical convergence is performed on the arch strain segments to generate a trend matrix.

[0009] The arch mileage data and the trend matrix are subjected to dual-gated edge weight reconstruction based on structural similarity and spatiotemporal distance to generate edge weight matrix data. Then, the trend matrix is ​​subjected to neighborhood consistency convergence based on the edge weight matrix data to generate a common mode trend sequence.

[0010] Perform differential separation processing on the common mode trend sequence and the trend matrix to generate a differential mode trend sequence. Perform on-graph inflection point consensus screening on the differential mode trend sequence and the arch strain segment to generate a consensus inflection point set.

[0011] A dual-channel energy construction of differential deviation and inflection point consensus is performed on the differential trend sequence and the consensus inflection point set to generate an abnormal energy sequence;

[0012] The abnormal energy sequence, the edge weight matrix data, and the arch mileage data are subjected to on-map diffusion backtracking positioning and encapsulation to generate trend analysis results.

[0013] Secondly, this invention discloses a trend analysis system for long-term monitoring data of lightweight steel arches, comprising:

[0014] The data acquisition module is used to acquire the steel arch strain sequence and arch position mileage data of the target object;

[0015] The trend feature aggregation module is used to perform arch position mapping and segmentation processing on the steel arch strain sequence based on the arch position mileage data, generate arch strain segments, and perform symmetrical aggregation on the arch strain segments to generate a trend matrix.

[0016] The common mode trend analysis module is used to perform dual-gated edge weight reconstruction of the arch mileage data and the trend matrix based on structural similarity and spatiotemporal distance, generate edge weight matrix data, and perform neighborhood consistency convergence on the trend matrix based on the edge weight matrix data to generate a common mode trend sequence.

[0017] The inflection point identification module is used to perform differential separation processing on the common mode trend sequence and the trend matrix to generate a differential mode trend sequence, and to perform on-graph inflection point consensus screening on the differential mode trend sequence and the arch strain segment to generate a consensus inflection point set.

[0018] An abnormal energy analysis module is used to perform dual-channel energy construction of differential mode deviation and inflection point consensus on the differential mode trend sequence and the consensus inflection point set to generate an abnormal energy sequence.

[0019] The trend analysis result output module is used to perform on-map diffusion backtracking positioning and encapsulation on the abnormal energy sequence, the edge weight matrix data, and the arch mileage data to generate trend analysis results.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] 1. This scheme generates a weighted matrix by performing dual-gated edge weight reconstruction on the arch mileage data and trend matrix, and generates a common-mode trend sequence through neighborhood consensus convergence. This achieves the weakening of long-distance accidental similarity and the suppression of short-distance morphological differences, so that regional common convergence is centrally expressed. The common-mode trend sequence and trend matrix are differentially separated to generate a differential-mode trend sequence, and a consensus inflection point set is generated through graph inflection point consensus screening. This achieves the removal of regional covariance and the filtering of false inflection points through neighborhood mutual verification, making local abnormal change points more reliable. The differential-mode trend sequence and consensus inflection point set are used to construct an abnormal energy sequence through dual-channel energy construction, so as to unify continuous deviation and discrete mutation into a diffusible signal and maintain the peak gradient.

[0022] 2. This scheme measures the difference between the mean values ​​of the front and rear half windows of the arch strain segment according to the window segment index table to generate morphological offset. It then maps the morphological offset to the mutation degree to generate morphological matching data. It introduces original strain morphological evidence to verify the consistency of the differential mode mutation direction, so that pseudo-mutations caused only by noise or isolated anomalies are significantly weakened after the consistent mapping. It performs weighted fusion of mutation degree, similarity matrix and morphological matching data to generate consensus strength data and performs maximum transition truncation screening to generate a consensus inflection point set. It uses the mutual verification weighting of structural similar neighborhoods to improve the spatial consistency of effective inflection points and automatically separate strong and weak inflection points, making the output inflection points more stable and the positioning more reliable. Attached Figure Description

[0023] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0024] Figure 1 A flowchart illustrating the steps of the trend analysis method for long-term monitoring data of lightweight steel arches provided by this invention.

[0025] Figure 2 This is a schematic diagram of the process for generating edge weight matrix data provided by the present invention;

[0026] Figure 3 A schematic diagram of the process for generating a consensus inflection point set provided by the present invention;

[0027] Figure 4 A schematic diagram of the process for generating abnormal energy sequences provided by the present invention;

[0028] Figure 5 A schematic diagram of the module functions of the trend analysis system for long-term monitoring data of lightweight steel arches provided by the present invention. Detailed Implementation

[0029] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0030] Application Overview:

[0031] During the long-term monitoring of lightweight steel arch groups in underground salt mine transport tunnels, factors such as surrounding rock creep, hydrothermal coupling, and brine erosion lead to the accumulation of tunnel deformation and the evolution of the stress on the support structure. This poses challenges to the reliability and interpretability of trend analysis of monitoring data. Existing technologies use fixed spatial adjacency relationships to aggregate trends of adjacent arch positions in order to suppress single-point noise and enhance the spatial continuity of anomaly location. However, in the complex underground environment of dynamically changing surrounding rock conditions, construction differences, and stress transmission evolution, the trends of adjacent arch positions show significant differentiation at different times. Fixed aggregation methods forcibly mix dissimilar neighborhoods, causing real anomalies to be smoothed out and producing pseudo-consistency. As a result, inflection point detection based on aggregation results relies excessively on the amplitude of single-sequence mutations. Under the influence of salt spray and dust peak interference, data noise, and changes in sampling intervals, pseudo-inflection points are easily generated. Furthermore, the lack of an arch group consensus verification mechanism makes anomaly triggering unstable and makes it difficult to accurately characterize the propagation pattern of anomalies in the arch group, ultimately affecting the accuracy of trend analysis results.

[0032] For example, in the monitoring practice of a transport roadway in an underground salt mine, a group of lightweight steel arches was continuously deployed along the roadway. The mileage data of each arch was accurately recorded, and the coordinates of each arch position were recorded. In areas with uneven surrounding rock conditions, at the boundary between local soft rock and hard rock sections, due to differences in the creep rate of the surrounding rock and the degree of brine erosion, the strain trends of adjacent arch positions diverged significantly in the short term: one arch position showed a continuously accelerating deformation trend due to softening of the surrounding rock, while adjacent arch positions showed a stable trend due to support adjustments. Existing fixed aggregation methods forcibly smoothed these trends, causing the true deformation anomalies to be masked. At the same time, random noise in stable arch positions was misjudged as anomalies, forming pseudo-consistency. As a result, the monitoring system incorrectly identified stable areas as high-risk areas, while the true deformation areas were not promptly warned, affecting the timeliness and accuracy of the roadway safety status assessment.

[0033] If the above problems are not resolved, the inaccuracy of trend analysis results will persist, making it difficult to reliably identify real structural anomalies, potentially delaying critical maintenance opportunities and increasing the risk of roadway instability. At the same time, the lack of characterization of anomaly propagation patterns makes it difficult to assess the overall performance of the support system, which may lead to unnecessary maintenance operations or overlook potential hazards, ultimately threatening the stability of the mine's safe production system and personnel safety.

[0034] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0035] Example 1:

[0036] Please see Figure 1 A trend analysis method for long-term monitoring data of lightweight steel arches, applied to the trend analysis of continuously deployed lightweight steel arch groups in underground salt mine transportation tunnels, includes the following steps:

[0037] Obtain the strain sequence and arch position mileage data of the target object's steel arch;

[0038] Based on the arch position mileage data, the steel arch strain sequence is subjected to arch position mapping and segmentation processing to generate arch-specific strain segments, and symmetrical convergence is performed on the arch-specific strain segments to generate a trend matrix.

[0039] The arch mileage data and trend matrix are reconstructed using a dual-gated weighted approach based on structural similarity and spatiotemporal distance to generate weighted matrix data. Then, neighborhood consistency convergence is performed on the trend matrix based on the weighted matrix data to generate a common-mode trend sequence.

[0040] Perform differential separation processing on the common mode trend sequence and trend matrix to generate the differential mode trend sequence. Perform on-graph inflection point consensus screening on the differential mode trend sequence and the arch strain segment to generate a consensus inflection point set.

[0041] Perform dual-channel energy construction of differential deviation and inflection point consensus on the differential trend sequence and consensus inflection point set to generate an abnormal energy sequence;

[0042] Perform on-map diffusion backtracking and encapsulation on abnormal energy sequences, edge weight matrix data, and arch mileage data to generate trend analysis results.

[0043] Among them, the steel arch strain sequence refers to the set of structural strain data continuously collected and organized in chronological order on a lightweight steel arch that is continuously deployed in the underground salt mine transportation tunnel.

[0044] Arch position mileage data refers to a structured data object used to uniquely identify the spatial position of each lightweight steel arch in the longitudinal direction of the underground salt mine transportation tunnel.

[0045] The arch-strain segment refers to the strain time series corresponding to multiple lightweight steel arches continuously deployed in the underground salt mine transportation tunnel, which are mapped one by one according to the arch position number and mileage order. The strain data of each steel arch on the entire monitoring time axis are independently segmented and packaged to form a structured data unit of single arch - continuous time period - ordered strain value set.

[0046] A trend matrix is ​​a two-dimensional data object used to uniformly characterize the long-term changing trend of each lightweight steel arch within each time window.

[0047] Edge weight matrix data refers to a two-dimensional matrix data object used to describe the structural correlation strength between lightweight steel arches in underground salt mine transportation tunnels;

[0048] Common mode trend sequence refers to a time series data set that characterizes the regional common change trend of lightweight steel arch groups in underground salt mine transportation tunnels;

[0049] The differential trend sequence refers to the deviation expression sequence formed by differentially separating the local trend expressed by the trend matrix of each lightweight steel arch from the common trend of the arch group in the same window segment under a unified window segment system.

[0050] The consensus inflection point set refers to the set of inflection point data that meets the consistency requirements of the three types of evidence after unified quantification and fusion of differential mode trend change evidence, original strain form evidence, and graph structure neighborhood mutual evidence under the same window segment index system.

[0051] Anomaly energy sequences refer to energy-based trend representation data used to perform diffusion and backtracking localization on graph structures;

[0052] Trend analysis results refer to the structured output data objects after long-term trend analysis of the lightweight steel arch group continuously deployed in the underground salt mine transportation tunnel.

[0053] This scheme generates arch strain segments by performing arch position mapping on the steel arch strain sequence based on arch position mileage data, and then symmetrically converges these segments to generate a trend matrix. This can suppress the dominant influence of salt dust spikes and occasional sampling disturbances on the trend, highlight the structural shifts in the first and second halves of the same window segment, and thus stably preserve the long-term trend pattern. The scheme performs dual-gated edge weight reconstruction on the arch position mileage data and the trend matrix to generate an edge weight matrix, and then performs neighborhood consistency convergence to generate a common mode trend sequence. This can simultaneously suppress long-distance pseudo-similarity and short-distance heteromorphic interference, enabling the mutual verification neighborhood to adaptively update with the working conditions and extract the common change background of the arch group to avoid the overall convergence from masking local anomalies. The scheme performs differential separation of the common mode trend sequence and the trend matrix to generate a differential mode trend sequence, and combines the arch strain segments to perform on-graph inflection point consensus screening to generate a consensus inflection point set. This can decouple local deviations from regional covariance, and use the original strain pattern and neighborhood voting as dual evidence to screen out pseudo-inflection points, improving the reliability and interpretability of inflection point triggering.

[0054] By constructing anomaly energy sequences from differential trend sequences and consensus inflection point sets using dual-channel energy, continuous deviations and discrete mutations can be unified into propagable signals. This strengthens mutation windows with collective consensus and reduces isolated noise energy, thereby improving the resolution of anomaly source localization. The anomalous energy sequences, edge weight matrix data, and arch mileage data are subjected to diffusion backtracking localization and encapsulation to generate trend analysis results. Under graph constraints, spatial gradients can be formed and backtracking can be used to lock the source arch and diffusion range, achieving stable localization conclusions for anomaly locations and influence intervals under sparse sensing conditions.

[0055] The above describes a complete scheme for trend analysis of long-term monitoring data of lightweight steel arches. The following section details the acquisition of the steel arch strain sequence and arch location mileage data for the target object, including:

[0056] The strain sequence of the steel arch of the target object is obtained by strain sensors; the strain sequence of the steel arch includes, but is not limited to, arch position number, acquisition time point and strain value at the corresponding time point;

[0057] The arch mileage data of the target object is obtained through mileage calibration equipment; the arch mileage data includes, but is not limited to, the arch number and the corresponding mileage value.

[0058] Based on the arch position mileage data, the steel arch strain sequence is subjected to arch position mapping and segmentation to generate arch-specific strain segments:

[0059] The arch position mileage data is sorted sequentially, and all arch position identifiers are arranged in ascending order of mileage value to generate a mileage sorting table. The calculation process involves sorting the mileage value corresponding to each arch position identifier. The first position after sorting corresponds to the minimum mileage value, and the last position corresponds to the maximum mileage value, thus obtaining the correspondence between sorting position, arch position identifier, and mileage value.

[0060] The strain sequence of the steel arch is processed by extracting the arch position identifier from each strain record while maintaining the original time order, and generating a strain identifier column. The calculation process is to arrange the arch position identifier corresponding to each record in the order of the records in the strain sequence of the steel arch to form an identifier sequence.

[0061] Based on the arch position identifiers in the mileage sorting table, position correspondence calculations are performed on the strain identifier column to generate a mapping position table. The calculation process is as follows: for each arch position identifier, all position numbers appearing in the strain identifier column are counted, and all position numbers corresponding to the same arch position identifier are arranged in ascending order to obtain the corresponding result of the arch position identifier-position number set. Then, based on the mapping position table, continuous merging processing is performed to generate segment index groups. The calculation process is as follows: adjacent positions in the position number set corresponding to the same arch position identifier are combined in continuous order, and the first position number in each group is used as the segment start position and the last position number is used as the segment end position to calculate the segment start position and segment end position, thus forming the segment index group corresponding to the arch position.

[0062] Based on the segment index group, the strain record set corresponding to the start position and end position of each segment is extracted from the strain sequence of the steel arch to generate the arch strain segment. The calculation process is to extract all strain records from the start position to the end position of the segment in each segment index group, and collect all the extracted results corresponding to the same arch position identifier to form a data object of arch position identifier-strain time period set, that is, the arch strain segment.

[0063] The above describes how to obtain the strain sequence and arch location mileage data of the target object's steel arch. The following describes how to perform symmetrical convergence on the strain segments of the arch to generate a trend matrix, specifically including:

[0064] The arch strain segments are uniformly segmented into windows to generate a window index table. Based on the window index table, the arch strain segments are separated into front and back halves and rearranged in reverse order to generate a symmetrical pairing sequence.

[0065] The symmetrical paired sequences are subjected to pairwise difference and difference mean aggregation to generate a symmetrical difference set. The symmetrical difference set is then arranged in two dimensions according to the arch index and window segment index to generate a trend matrix.

[0066] Among them, the window segment index table refers to the data mapping structure generated by performing equal-scale continuous segmentation processing on all arch strain time series according to a unified time base;

[0067] Symmetric pairing sequence refers to a time-symmetric structured data object constructed for the strain data of each arch within the same window segment;

[0068] A symmetric difference set refers to a data object formed within the same window segment by symmetrical pairing and pairwise difference processing of the arch strain segments, used to characterize the structural offset features before and after time within the window segment.

[0069] The above content will be described in detail below:

[0070] The arch-strain segments are uniformly segmented using windowing to generate a windowing index table. Based on this index table, the arch-strain segments are then separated into front and rear halves and rearranged in reverse order to generate a symmetrical pairing sequence.

[0071] The strain time corresponding to each arch position in the arch strain segment is corrected for time sequence, so that all strain values ​​under the same arch position form a continuous sequence according to the sampling order. Then, the difference between adjacent sampling times in the continuous sequence of each arch position is obtained to obtain the set of adjacent time differences. The sum of the set of adjacent time differences is divided by the number of time differences to obtain a unified reference interval. Then, the earliest sampling time and the latest sampling time in the continuous sequence of all arch positions are extracted, and the difference between the two is obtained to obtain the total duration. The total duration is divided by the reference interval to obtain the total step size. On this basis, the total step size is divided equally, and each continuous step size interval is corresponding to a window segment. Each window segment is given a window segment number, start step position and end step position, thus generating a window segment index table. Each index record in the window segment index table is uniquely represented by the window segment number and the corresponding step position range.

[0072] Based on the window segment index table, strain subsequences corresponding to each window segment number are extracted from the continuous sequence of each arch position to obtain windowed strain segments. Then, a front and back half separation process is performed on each windowed strain segment. Specifically, the total number of sampling points contained in the windowed strain segment is counted first, and the total number of sampling points is divided by two to obtain the midpoint position. The sampling points before the midpoint position are arranged in order to form the front half sequence, and the sampling points after the midpoint position are arranged in order to form the back half sequence. Then, a reverse rearrangement process is performed on the back half sequence. Specifically, the last sampling point in the back half sequence is adjusted to the first position, and the second to last sampling point is adjusted to the second position, until all positions are reversed to generate a reverse back half sequence. Then, the front half sequence and the reverse back half sequence are combined one by one according to the corresponding positions to obtain several paired units. Each paired unit consists of a front half sampling value and a reverse back half sampling value. All paired units in the same window segment are arranged in the original position order to generate the symmetrical paired sequence corresponding to the window segment.

[0073] The symmetrical paired sequences are subjected to pairwise differencing and mean difference aggregation to generate a symmetrical difference set. This symmetrical difference set is then arranged in two dimensions according to the arch index and window segment index to generate a trend matrix.

[0074] For symmetrical pairing sequences, a pairwise difference processing is performed. Specifically, for each pair of values, the system subtracts the first half of the sampled value from the reversed second half of the sampled value in the group to obtain the single difference value corresponding to the pair. This converts each symmetrical relationship into a directional difference result. After all single differences are generated, the difference mean aggregation processing is performed on all single differences within the same arch and window segment according to the relationship between the arch and the window segment. Specifically, the calculation process is as follows: first, all single differences within the same arch and window segment are accumulated to obtain the sum of differences; then, the sum of differences is divided by the number of pairing groups within the same arch and window segment to obtain the mean difference result corresponding to the same arch and window segment, and this mean difference result is defined as the symmetrical difference.

[0075] After repeatedly performing the above continuous processing of pairwise difference, difference sum formation, and difference mean aggregation on all arch positions and all window segments, a symmetrical difference set consisting of multiple symmetrical differences is generated. Each symmetrical difference in the symmetrical difference set corresponds to a unique arch position index and a unique window segment index. After the symmetrical difference set is generated, a two-dimensional positioning and arrangement process is further performed on the symmetrical difference set. The specific processing method is as follows: using the arch position index as the row positioning basis and the window segment index as the column positioning basis, each symmetrical difference is filled into the two-dimensional position corresponding to its arch position index and window segment index, thereby forming a matrix data structure organized by rows to represent arch positions and columns to represent window segments. When all symmetrical differences have been filled into their corresponding positions, a trend matrix is ​​output.

[0076] This scheme generates a window index table by uniformly segmenting the strain sections of the arches, and then performs front and rear half separation and reverse rearrangement of the rear half to generate symmetrical paired sequences. This ensures that the strain data within the same window segment of the same arch are symmetrically aligned on the time scale, eliminating the comparison bias caused by inconsistent sampling start and end, and structurally separating slow drift and instantaneous spikes from the mixed sequence. The symmetrical paired sequences are differentially divided pair by pair, and the difference mean convergence is performed on the difference results to generate a symmetrical difference set. The paired difference is used to offset the common mode fluctuations within the same window segment and highlight the structural offset of the front and rear half windows. The impact of isolated high values ​​caused by salt dust disturbance on the trend expression is reduced by mean convergence, and the offset direction remains discernible. The symmetrical difference set is arranged in two dimensions according to the arch index and window segment index to generate a trend matrix. The offset intensity of each arch position is uniformly expressed in a matrix form comparable to that of the same window, so that the subsequent similarity calculation and neighborhood voting on the arch group map have a consistent data alignment benchmark and improve the spatial resolution of anomaly location.

[0077] The above describes the symmetrical convergence of strain segments in the arch section to generate a trend matrix. The following describes the dual-gated edge-weight reconstruction of the arch position mileage data and the trend matrix using structural similarity and spatiotemporal distance to generate edge-weight matrix data. Please refer to [reference needed]. Figure 2 , Figure 2This is a schematic diagram of the process for generating edge weight matrix data provided in an embodiment of this application. Generating edge weight matrix data specifically includes:

[0078] The distance-gated data is generated by performing pairwise difference calculations and reciprocal decay mapping on the arch mileage data.

[0079] Under the constraints of the window index table, the trend matrix is ​​extracted and rearranged within the same window, and then similarity mapping is performed by combining the symmetric difference set to generate a similarity matrix.

[0080] The distance-gated data and the similarity matrix are subjected to mutual attraction-gated fusion, and the mutual attraction-gated fusion result is multiplied and fused with the similarity matrix to generate edge weight matrix data.

[0081] Among them, distance-gated data refers to a two-dimensional matrix data object used to characterize the propagation constraint strength between any two steel arches in the spatial dimension;

[0082] A similarity matrix is ​​a structural similarity matrix generated by mapping the trend morphological differences between different arches within the same window segment under the constraints of a window segment index table.

[0083] The mutual gating fusion result refers to the gating coefficient matrix formed by bidirectionally constraining and correcting distance gating data and similarity matrix under a unified arch position index within the same window segment.

[0084] The above content will be described in detail below:

[0085] The distance-gated data is generated by performing pairwise difference calculations and inverse attenuation mapping on the arch mileage data. The specific calculation formula is as follows:

[0086] ;

[0087] In the formula, Indicates the first Arch mileage data corresponding to each arch position. Indicates the first The mileage data corresponding to each arch position is normalized during the calculation.

[0088] Under the constraints of the window index table, the trend matrix is ​​extracted and rearranged within the same window, and then similarity mapping is performed using a symmetric difference set to generate a similarity matrix:

[0089] Based on the window segment index table, the trend matrix is ​​subjected to same-window extraction. Specifically, all symmetrical differences of arches within the same window segment are extracted from the trend matrix according to the same window segment number. The extracted symmetrical differences are then rearranged according to the arch order to generate the same-window trend group for that window segment. Next, all symmetrical differences with the same window segment number are extracted from the symmetrical difference set and rearranged according to the same arch order as the same-window trend group to generate the same-window difference group for that window segment. Subsequently, similarity mapping is performed on any two arches within the same window segment. The specific calculation formula is as follows:

[0090] ;

[0091] In the formula, Indicates in window segment inside, arch position With arch similarity value, Indicates in window segment within, no. The symmetrical difference corresponding to each arch position Indicates in window segment within, no. The symmetrical differences corresponding to each arch position, and the above data have all been normalized during the calculation;

[0092] Complete the pairwise similarity mapping between all arch positions within the same window segment in the same manner as described above, and arrange all the obtained similarity values ​​according to the row and column positions of the arch positions to form the window segment similarity table for that window segment. Then, summarize and arrange the window segment similarity tables corresponding to each window segment in order of window segment number to finally generate a similarity matrix.

[0093] The distance-gated data and the similarity matrix are subjected to cross-referencing gating fusion, and the cross-referencing gating fusion result is multiplied and fused with the similarity matrix to generate edge weight matrix data:

[0094] The distance-gated data and the similarity matrix are aligned with the corresponding arch positions to establish a same-dimensional matrix pair. The first gating operation is performed on each alignment position. The calculation process is to take the distance-gated data and the similarity value at that position to obtain the first gating value, thereby realizing the attenuation constraint of the similarity intensity by the distance gating. Then, the first gating value and the distance-gated data are subjected to a second gating operation. The calculation process is to take the first gating value and the distance-gated data at that position to obtain the second gating value, thereby realizing the reverse correction of the distance gating intensity by the constrained similarity intensity. The set of all alignment positions of the second gating value is the mutual gating fusion result.

[0095] The mutual gating fusion result is multiplied with the similarity matrix position by position. The calculation process is to take the mutual gating fusion result at each alignment position and multiply it with the similarity value to obtain the edge weight. All edge weights are then arranged in matrix form according to the arch position pairs and output as edge weight matrix data.

[0096] Based on the edge weight matrix data, perform neighborhood consistency aggregation on the trend matrix to generate a common mode trend sequence:

[0097] The edge weight matrix data is subjected to row-direction normalization. The specific process is as follows: sum all edge weights in each row of the edge weight matrix data to obtain the total weight of the corresponding arch position, and then divide each edge weight in that row by the total weight to generate the normalized neighborhood weight corresponding to the original edge weight matrix rows and columns, so that the proportion of the effect of each arch position on all its neighboring arch positions is unified to the same scale.

[0098] The trend matrix undergoes window segmentation processing. Specifically, the process involves extracting the symmetrical differences of all monitored arch positions within the same window segment from the trend matrix, window by window, to form the trend column group corresponding to that window segment. The normalized neighborhood weights are then multiplied with the trend column group, ensuring that the neighborhood convergence value for each arch position is equal to the sum of the normalized neighborhood weights of that arch position multiplied by the symmetrical differences of the corresponding arch positions in the trend column group. This generates the neighborhood convergence vector for all arch positions within that window segment. Furthermore, consistency compression processing is applied to the neighborhood convergence vector. The specific processing steps are as follows: First, sum all the convergence values ​​in the neighborhood convergence vector to obtain the total convergence amount. Then, divide the total convergence amount by the number of arches participating in the convergence to generate the common mode symmetric difference of the window segment. This common mode symmetric difference is used to characterize the overall change direction and overall change level of each monitored arch under the constraint of the edge weight matrix data. Repeat the window segment separation processing, neighborhood consistency convergence processing, and consistency compression processing in the order of all window segments. Arrange the common mode symmetric differences generated by each window segment in sequence to finally generate the common mode trend sequence.

[0099] Perform difference separation on the common-mode trend sequence and the trend matrix to generate the differential-mode trend sequence:

[0100] The number of columns in the trend matrix is ​​read as the total number of time windows, the sequence length of the common mode trend sequence is read as the number of common mode terms, and the common mode trend sequence is expanded into a single row according to the time window order. Then, it is copied and expanded along the steel arch dimension to generate a common mode mapping matrix with the same number of rows and columns as the trend matrix. The value at any position in the common mode mapping matrix is ​​taken from the common mode trend sequence value corresponding to the column number at that position.

[0101] The trend matrix and the common mode mapping matrix are subjected to difference separation processing. Specifically, at the same row and column number, the common mode value in the common mode mapping matrix is ​​subtracted from the symmetric difference in the trend matrix to obtain the difference value at that position. The difference values ​​at all positions are organized in the original matrix arrangement to generate a difference matrix. Each element in the difference matrix represents the deviation of the corresponding steel arch from the overall common trend under the corresponding time window. On this basis, the difference matrix is ​​further subjected to serialization and sorting processing. Specifically, the difference values ​​of each row are extracted according to the steel arch number order, and the difference values ​​of each row are connected according to the time window order to form the local deviation trend subsequence of the corresponding steel arch. Then, all local deviation trend subsequences are combined according to the steel arch number order to generate the differential mode trend sequence.

[0102] This scheme generates distance-gated data by performing pairwise difference calculations and reciprocal decay mapping on arch mileage data. This quantifies the spatial distance between arches as a continuous gating strength and automatically weakens long-distance correlations, thereby suppressing false strong edges caused by accidental synchronization of non-adjacent arches and stabilizing the mutual evidence neighborhood range. Under the constraint of the window segment index table, the trend matrix is ​​extracted and rearranged within the same window and then combined with the symmetric difference set for similarity mapping to generate a similarity matrix. This ensures that structural similarity is calculated only within the same window segment and superimposed with symmetric offset evidence to strengthen the identification of slow drift and structural deviation. The identification of similarity matrices reduces false similarities caused by cross-window mixing and spike disturbances, and improves the comparability and noise resistance of similarity. The distance-gated data and similarity matrix are mutually fused, and the mutual fusion result is multiplied and fused with the similarity matrix to generate edge weight matrix data. This allows spatial constraints and structural similarity to mutually correct each other and jointly determine edge strength, thereby suppressing both long-distance even similarity and short-distance heteromorphic false connections. This improves the accuracy of graph structure in depicting real mutual verification relationships and provides a more reliable propagation path for subsequent common mode and differential mode separation and source arch backtracking.

[0103] The above describes the dual-gated edge-weight reconstruction of the arch mileage data and trend matrix using structural similarity and spatiotemporal distance to generate edge-weight matrix data. The following describes the process of performing consensus filtering on the differential modulus trend sequence and the arch strain segments to generate a consensus inflection point set. Please refer to [reference needed]. Figure 3 , Figure 3 This is a schematic diagram of the process for generating a consensus inflection point set provided in an embodiment of this application. Generating the consensus inflection point set specifically includes:

[0104] The differential trend series is paired with adjacent windows and the differential magnitude is measured according to the window segment index table to generate a measure of the degree of abrupt change.

[0105] The difference between the mean values ​​of the front and rear half windows of the arch strain segment is measured according to the window segment index table to generate the shape offset. The shape offset and the abrupt change degree are mapped in a consistent manner to generate shape matching data.

[0106] The mutation degree, similarity matrix and morphological matching data are weighted and fused to generate consensus strength data. The consensus strength data is then filtered by maximum transition truncation to generate a consensus inflection point set.

[0107] Among them, the degree of mutation refers to a numerical data object used to characterize the magnitude of the difference in the differential mode between adjacent window segments for the same arch position;

[0108] Morphological offset refers to a numerical data object used to characterize the structural change amplitude of the original strain time series of the same arch within the same window segment in the time dimension between the preceding and following stages.

[0109] Morphology matching data refers to the quantitative result formed by numerically mapping the original strain morphology offset trend and differential mode abrupt change trend of the same arch position within the same window segment;

[0110] Consensus strength data refers to a unified numerical data object used to characterize the overall consistency of three types of evidence: differential mode mutation evidence, original morphological consistency evidence, and structural similarity mutual evidence at a certain arch position within a certain window segment.

[0111] The above content will be described in detail below:

[0112] The differential trend series is paired with adjacent windows and the difference magnitude is measured according to the window segment index table to generate a measure of the degree of abrupt change.

[0113] Based on the order of window segment numbers in the window segment index table, the differential trend sequence values ​​of the same arch position on two adjacent window segments are paired to output adjacent window differential mode pairs. The process of obtaining adjacent window differential mode pairs is as follows: the differential trend sequence value of the first window segment is paired with the differential trend sequence value of the second window segment whose window segment number is immediately following it. Then, the second window segment is paired with the third window segment to form the next pair. All adjacent windows are paired in the order of window segment numbers to obtain adjacent window differential mode pairs organized according to time adjacency.

[0114] For each adjacent window difference mode pair, a difference operation is performed to output the difference mode difference value. The calculation process of the difference mode difference value is as follows: the difference mode trend sequence value of the previous window segment is subtracted from the difference mode trend sequence value of the subsequent window segment to obtain the corresponding difference data, which is used to characterize the direction and magnitude of the difference mode change between adjacent windows. After obtaining the difference mode difference value, the amplitude measurement processing is performed on the difference mode difference value to output the abrupt change degree. The calculation process of the abrupt change degree is as follows: when the difference mode difference is a single-value data, the absolute value is directly taken to obtain the corresponding abrupt change degree. When the difference mode difference is a multi-component data, each component is squared first, then all squared results are summed, and then the square root of the summation result is obtained to obtain the corresponding abrupt change degree, so as to ensure that the generated data uniformly characterizes the comprehensive amplitude of the difference mode change between adjacent windows.

[0115] The difference between the mean values ​​of the front and rear half windows of the arch strain segment is measured according to the window segment index table to generate morphological offset. The morphological offset is then mapped to the abrupt change level to generate morphological matching data.

[0116] According to the window segment index table, the strain segment corresponding to each arch position is positioned in the window segment, the strain sub-segment in the current window segment is extracted, and the strain sub-segment is divided into the first half window strain group and the second half window strain group according to the window segment position. The first half window strain group corresponds to all strain values ​​between the start position and the middle position of the window segment, and the second half window strain group corresponds to all strain values ​​between the middle position and the end position of the window segment.

[0117] Mean values ​​were measured for the first and second half-window strain groups respectively to obtain the first half mean and the second half mean. The calculation process is as follows: sum all strain values ​​in the first half-window strain group and divide by the number of strain values ​​in the first half-window strain group to obtain the first half mean; sum all strain values ​​in the second half-window strain group and divide by the number of strain values ​​in the second half-window strain group to obtain the second half mean.

[0118] Perform a difference operation between the second half mean and the first half mean to obtain the morphological offset. The calculation process is as follows: subtract the first half mean from the second half mean. The result is used to characterize the overall offset direction and offset intensity of the strain morphology from the first half window to the second half window within the current window segment.

[0119] A consistent mapping process is performed on the morphological offset and the degree of mutation. Specifically, the absolute value of the morphological offset is first calculated to obtain the offset amplitude. Then, the difference between the offset amplitude and the degree of mutation is measured to obtain the amplitude difference. The calculation process is as follows: the absolute value of the offset amplitude minus the degree of mutation is taken. Subsequently, a consistency coefficient is generated based on the offset amplitude, the degree of mutation, and the amplitude difference. The calculation process is as follows: the sum of the offset amplitude and the degree of mutation minus the amplitude difference is used as the numerator, and the sum of the offset amplitude and the degree of mutation plus one is used as the denominator. A division operation is performed to obtain the consistency coefficient, so that the consistency coefficient continuously represents the closeness of the morphological offset and the degree of mutation in amplitude. Further, a direction-preserving mapping is performed on the morphological offset to obtain the vector-preserving mapping. The calculation process is as follows: the morphological offset is used as the numerator, and the offset amplitude plus one is used as the denominator. A division operation is performed to obtain the vector-preserving mapping, so that the numerical range is compressed while preserving the offset direction. Finally, the consistency coefficient and the vector-preserving mapping are multiplied to output the morphological matching data.

[0120] The mutation degree, similarity matrix, and morphological matching data are weighted and fused to generate consensus strength data. Then, the consensus strength data is filtered by maximum transition truncation to generate a set of consensus inflection points.

[0121] Amplitude mapping is performed on the mutation severity values ​​to generate mutation normalization values. The calculation process of amplitude mapping is to add each mutation severity value to one to obtain the denominator, and then divide the mutation severity value by the denominator to obtain the mutation normalization value.

[0122] Neighborhood aggregation is performed on similar matrices to generate similarity aggregation values. The calculation process of neighborhood aggregation is to sum each row of the similar matrix within the same window segment to obtain the row sum, and then divide the row sum by the number of elements in that row to obtain the similarity aggregation value.

[0123] A consistent mapping is performed on the morphological matching data to generate a morphological normalization value. The calculation process of the consistent mapping is to add each morphological matching data to one to obtain the denominator, and then divide the morphological matching data by the denominator to obtain the morphological normalization value.

[0124] A fusion weight vector is generated, and weighted fusion is performed on the three types of normalized values ​​to obtain consensus strength data. The calculation process of the fusion weight vector is as follows: the mutation normalized value, similarity convergence value, and morphological normalized value are independently summed to obtain three weight base quantities. Then, each weight base quantity is divided by the sum of the three weight base quantities to obtain the corresponding fusion weight. Subsequently, the mutation normalized value is multiplied by its corresponding fusion weight to obtain the first weighting term, the similarity convergence value is multiplied by its corresponding fusion weight to obtain the second weighting term, and the morphological normalized value is multiplied by its corresponding fusion weight to obtain the third weighting term. Finally, the first weighting term, the second weighting term, and the third weighting term are added together to obtain the consensus strength data.

[0125] The consensus strength data is subjected to maximum transition truncation filtering to generate a consensus inflection point set, and the calculation process is as follows: Within the same window segment, the consensus strength data of all arch positions are sorted from largest to smallest to obtain a sorted sequence. The difference operation is performed on adjacent items in the sorted sequence to obtain a difference sequence. The calculation process of the difference operation is to subtract the next item from the previous item to obtain the difference value and take its absolute value. The item with the largest difference value in the difference sequence is located as the maximum transition position, and the first segment of the sorted sequence corresponding to the maximum transition position is used as the truncated retention segment. Finally, the arch position number and window segment number corresponding to each item in the truncated retention segment are output as the consensus inflection point set.

[0126] This scheme generates a mutation degree quantity by pairing adjacent windows and measuring differential amplitudes according to the window segment index table for the differential mode trend sequence. This transforms local deviations from vector changes into mutation intensity quantities comparable within the same window, suppressing pseudo-mutations caused by cross-window mismatches and improving the stability of inflection point candidates. It also measures the difference between the mean values ​​of the front and rear half windows for the arched strain segment according to the window segment index table, generating a morphological offset quantity. This morphological offset quantity is then mapped consistently with the mutation degree quantity to generate morphological matching data. The original strain morphology is used to perform dual-evidence verification of differential mode mutations, weakening amplitude mutations caused by noise spikes due to a lack of morphological consistency. This improves the interpretability and anti-interference capability of true inflection points. The mutation degree quantity, similarity matrix, and morphological matching data are weighted and fused to generate consensus strength data. Maximum transition truncation is then performed on the consensus strength data to generate a consensus inflection point set. Neighborhood similarity weighting is introduced to form graph-based consensus, making it difficult for isolated outliers to obtain high consensus and resulting in their truncation and removal. This enables reliable screening and spatial positioning of true inflection points under sparse sensing conditions.

[0127] The above describes the process of performing consensus filtering on the differential mode trend sequence and the arching strain segment to generate a consensus inflection point set. The following describes the process of performing dual-channel energy construction on the differential mode deviation and inflection point consensus on the differential mode trend sequence and the consensus inflection point set to generate an abnormal energy sequence. Please refer to [reference needed]. Figure 4 , Figure 4 This is a schematic diagram of the process for generating anomaly energy sequences provided in an embodiment of this application. Generating anomaly energy sequences specifically includes:

[0128] Perform a scale-unified mapping on the differential trend series and the differential normalized quantity to generate comparable deviation data;

[0129] Among them, the differential mode normalization quantity is obtained by performing a proximity measurement on the common mode trend sequence and the trend matrix to generate an explanatory ratio value, and then performing normalization fusion on the explanatory ratio value and the differential mode trend sequence;

[0130] The gating and fusion results, consensus inflection point set and morphological matching data are subjected to gating and consistency reinforcement mapping to generate consistent reinforcement data. The consistent reinforcement data and comparable deviation data are then cross-linked and synthesized to generate anomaly energy sequences.

[0131] Among them, the differential mode normalization quantity refers to a unified scale modulation data object constructed to address the problem that the differential mode deviation amplitudes between different window segments and different steel arches cannot be directly compared;

[0132] Comparable deviation data refers to numerical data objects that characterize the magnitude deviation of the differential mode for each arch position and each window segment under a uniform scale;

[0133] The scale value refers to a proportional numerical data object used to characterize how closely the common modulus expression fits the overall trend matrix;

[0134] Consistency-enhanced data refers to a unified numerical representation data object used to enhance the structural consistency of consensus inflection points.

[0135] The above content will be described in detail below:

[0136] Perform a scale-unified mapping on the differential trend series and the differential normalized quantity to generate comparable deviation data. The specific calculation formula is as follows:

[0137] ;

[0138] In the formula, Indicates in window segment inside, arch position Differential trend series data, Indicates in window segment The normalized quantity of the difference modulus within, This represents the L2 norm. All the data above have been normalized during the calculation.

[0139] The differential modulus normalization factor is obtained by performing a proximity measurement on the common modulus trend sequence and the trend matrix to generate an explanatory ratio value, and then performing normalization fusion on the explanatory ratio value and the differential modulus trend sequence.

[0140] The common mode trend sequence and the trend matrix are aligned according to the same window segment order. The common mode trend sequence is used to perform a proximity measurement on the trend matrix. Specifically, the common mode trend sequence is expanded by window segment to form an aligned common mode table, so that the aligned common mode table and the trend matrix are consistent in row and column positions. Then, the trend matrix and the aligned common mode table are subtracted at corresponding positions to generate a deviation matrix. The deviation matrix is ​​used to characterize the overall deviation of the trend matrix from the common mode trend sequence.

[0141] The system performs absolute value aggregation on the deviation matrix to generate the approximate total. The calculation process involves taking the absolute values ​​of all positions in the deviation matrix and summing them. At the same time, the system performs absolute value aggregation on the trend matrix to generate the ontological total. The calculation process involves taking the absolute values ​​of all positions in the trend matrix and summing them. After obtaining the approximate total and the ontological total, the system performs proportionalization on the two to generate the explanatory proportional value. The calculation process involves dividing the ontological total by the sum of the ontological total and the approximate total. The result is the explanatory proportional value.

[0142] Normalization and fusion processing is performed on the explanatory ratio value and the differential trend sequence to generate the differential normalization quantity. Specifically, the absolute value aggregation is first performed on the differential trend sequence to generate the differential total. The calculation process is to take the absolute value of the values ​​at all positions in the differential trend sequence and then sum them. Then, the differential total and the explanatory ratio value are summed to generate the normalization base quantity. Finally, the values ​​at each position in the differential trend sequence are divided by the normalization base quantity to obtain the differential normalization quantity.

[0143] The gating and fusion results, consensus inflection point set, and morphological matching data are subjected to gating and consistency reinforcement mapping to generate consistent reinforcement data. Then, the consistent reinforcement data and comparable deviation data are cross-referenced and synthesized to generate anomaly energy sequences.

[0144] The mutual referencing gating fusion results are mapped to the same scale to generate gating base values. Specifically, the mutual referencing gating fusion results corresponding to all arch positions within the same window segment are summed to obtain the total gating value of the window segment. Then, the mutual referencing gating fusion result of each arch position is divided by the sum of the total gating value of the window segment and one to obtain the gating base value of the corresponding arch position. Subsequently, the consensus inflection point set is processed by position expansion to generate inflection point indication data. Specifically, a position table with the same dimension as the gating base value is constructed according to the arch position number and the window segment number. The arch position number - window segment number positions that have been recorded in the consensus inflection point set are assigned a value of one, and the remaining positions are assigned a value of zero.

[0145] Gated consistency enhancement mapping is performed on the gating base value, inflection point indication data and morphological matching data to generate consistent enhancement data. Specifically, the gating base value at the same position is multiplied by the morphological matching data to obtain the gate coupling value. Then, the gate coupling value is added to the inflection point indication data at the corresponding position to obtain the inflection point merge value. Subsequently, a compression operation is performed on the inflection point merge value at each position, which is divided by the sum of the inflection point merge value at that position and one, to obtain consistent enhancement data.

[0146] A cross-reference synthesis is performed on the consistent reinforcement data and the comparable deviation data to generate an anomalous energy sequence. The specific processing is as follows: First, for each position, calculate one minus the consistent reinforcement data and one minus the comparable deviation data, then multiply the two to obtain the joint suppression value. Then, subtract the joint suppression value from the one to obtain the anomalous energy value at that position. This operation method ensures that the anomalous energy value increases synchronously when either the consistent reinforcement data or the comparable deviation data increases, and the anomalous energy value is further enhanced when both increase simultaneously, thereby avoiding the value stacking distortion caused by simple addition. Finally, the anomalous energy values ​​corresponding to each arch position are concatenated in the order of the window segment numbers to output the anomalous energy sequence.

[0147] This scheme generates comparable deviation data by performing a scale-unified mapping on the differential mode trend sequence and the differential mode normalization quantity. It compresses the differential mode deviations of different window segments and different arch positions to the same comparable scale, avoiding the amplification or masking of local anomalies due to differences in amplitude scale. The differential mode normalization quantity is obtained by performing a proximity metric on the common mode trend sequence and trend matrix to generate an explanatory ratio value and then normalizing and fusing it with the differential mode trend sequence. The normalization scale is adaptively adjusted using the strength of regional covariance. When regional convergence dominates, it improves the resolution of local deviations and suppresses false deviations when noise dominates. The scheme performs gated consistency enhancement mapping on the mutual referencing fusion results, consensus inflection point set and morphological matching data to generate consistent enhancement data. It assigns higher energy weights to inflection point mutations that are reliable and morphologically consistent in the neighborhood and limits the enhancement of weakly gated isolated noise. The consistent enhancement data and comparable deviation data are mutually referred to synthesize to generate anomaly energy sequences. It unifies continuous deviations and discrete mutations into a single energy signal that can be diffused and traced back, making the source arch localization more stable and reducing false alarms under sparse sensing.

[0148] The above describes the dual-channel energy construction of differential deviation and inflection point consensus on the differential trend sequence and consensus inflection point set to generate anomaly energy sequences. The following describes the on-graph diffusion backtracking positioning and encapsulation of the anomaly energy sequences, edge weight matrix data, and arch mileage data to generate trend analysis results, specifically including:

[0149] Propagation constraint fusion is performed on the edge weight matrix data and the arch position mileage data to generate a diffusion kernel matrix. Then, diffusion operation is performed on the diffusion kernel matrix and the anomalous energy sequence to generate a diffusion field matrix.

[0150] Perform reverse backtracking matching on the diffusion field matrix and the explained proportion to generate backtracking matching data, and perform matching confidence modulation on the backtracking matching data and the mutation proportion to generate matching confidence data;

[0151] The mutation percentage is obtained by performing a percentage mapping on consistent reinforcement data and comparable deviation data.

[0152] Regional covariance suppression fusion is performed on the matched reliable data and the explained proportion values ​​to generate the source arch contribution sequence. Discrete measurement and encapsulation processing are then performed on the source arch contribution sequence to generate trend analysis results.

[0153] Among them, the diffusion kernel matrix refers to the standardized propagation kernel matrix obtained by fusing the structural connection strength and spatial distance constraints under the same window segment number, using the edge weight submatrix of the corresponding window segment in the edge weight matrix data as the basis for structural connection, and the mileage difference formed by the arch position mileage data as the propagation radius constraint.

[0154] The diffusion field matrix refers to a two-dimensional numerical matrix constructed using the arch position as the row index and the window segment as the column index in the same window segment coordinate system.

[0155] Retrospective matching data refers to a two-dimensional matching result data object used in a unified window coordinate system to characterize the closeness between the theoretical propagation form and the actual diffusion field form of each candidate source arch in each window segment;

[0156] The mutation proportion refers to the data object used to characterize the proportion of the mutation component driven by the inflection point consensus in the anomalous energy of a certain arch within a certain window segment, relative to the continuous deviation component.

[0157] The source arch contribution sequence refers to the ordered numerical sequence calculated for each candidate arch during the diffusion backtracking localization process on the graph, which is used to characterize the contribution intensity of the candidate arch as an anomalous source to the overall diffusion field.

[0158] The above content will be described in detail below:

[0159] The edge weight matrix data and the arch mileage data are fused using reciprocal decay to generate a diffusion kernel matrix. Then, the anomalous energy sequence and the diffusion kernel matrix are fused to generate a diffusion energy matrix.

[0160] The row and column positions of the edge weight matrix data are matched with the mileage values ​​of each arch in the arch mileage data according to the same arch numbering order, so that any row and column position points to the same pair of arches. Then, the mileage difference matrix between arches is calculated based on the arch mileage data. The calculation process is as follows: the difference between the mileage values ​​corresponding to any two arches is taken and the absolute value is obtained to obtain the mileage difference between the two arches. All the pairwise mileage differences are arranged according to the positional relationship of the source arch × the target arch to form the mileage difference matrix.

[0161] The mileage difference matrix is ​​subjected to reciprocal attenuation processing. The calculation process is as follows: each mileage difference value in the mileage difference matrix is ​​added to one and the reciprocal is taken to obtain the attenuation value corresponding to that position. This forms the reciprocal attenuation matrix. The smaller the mileage difference, the larger the corresponding reciprocal attenuation value, and the larger the mileage difference, the smaller the corresponding reciprocal attenuation value.

[0162] The edge weight matrix data and the reciprocal decay matrix are fused together. The calculation process is as follows: the values ​​at the same position of the two are multiplied one by one to obtain the kernel initial value matrix. The kernel initial value matrix is ​​then subjected to row normalization. The calculation process is as follows: the sum of the values ​​in each row of the kernel initial value matrix is ​​calculated, and each kernel initial value in that row is divided by the sum of the row to obtain the normalized propagation coefficient corresponding to that row, thus forming the diffusion kernel matrix.

[0163] The abnormal energy sequence is matrixed. Specifically, the abnormal energy of each arch position under the same window segment is arranged into an energy column vector according to the arch position order consistent with the diffusion kernel matrix. The energy column vectors of multiple window segments are then arranged column-wise to form an abnormal energy matrix. Subsequently, the system performs a fusion operation between the abnormal energy matrix and the diffusion kernel matrix. The calculation process is as follows: the diffusion kernel matrix is ​​multiplied by the abnormal energy matrix on the left, so that the abnormal energy of each source arch position under each window segment is redistributed to all target arch positions according to the propagation ratio in the diffusion kernel matrix, and the diffusion energy column vectors corresponding to each window segment are obtained. Then, the diffusion energy column vectors of all window segments are arranged column-wise to form a diffusion energy matrix.

[0164] Enhanced energy-driven peak-preserving fusion and splicing are performed on the diffusion energy matrix to generate the diffusion field matrix:

[0165] The gating fusion results are summed row by row and then divided by the total gating value of the same window segment to obtain the gating convergence amount. Then, the consensus strength data, gating convergence amount and inflection point indication data are multiplied one by one to obtain the reinforcement base amount. Then, the reinforcement energy data is obtained by dividing the reinforcement base amount by (1 + reinforcement base amount). Next, the energy synthesis amount is obtained by dividing 1 - (1 - comparable deviation intensity) × (1 - reinforcement energy data). The mutation proportion is obtained by dividing the reinforcement energy data by (1 + reinforcement energy data + comparable deviation intensity). Finally, the energy synthesis amount and the mutation proportion are combined into an energy grid table according to the arch position × window segment.

[0166] The energy synthesis values ​​of each window segment in the energy grid table are rearranged into a grid energy matrix with the same dimension as the diffusion energy matrix according to "arch position × window segment". Then, the enhanced energy data is used as the peak shape retention driving quantity. The fusion value is calculated for each arch position and each window segment. The calculation process of the fusion value is diffusion energy value × (1 - enhanced energy value) + grid energy value × enhanced energy value. Thus, when the enhanced energy value is large, more local peaks are retained in the fusion value, and when the enhanced energy value is small, more diffusion gradients are retained in the fusion value. Then, the fusion values ​​of all arch positions in the same window segment are spliced ​​together in the order of arch positions to form the field vector of that window segment. Then, all window segment field vectors are continuously spliced ​​together in the order of window segments to obtain the diffusion field matrix.

[0167] Perform a backtracking match on the diffusion field matrix and the explained scale values, including window suppression mapping and in-window total normalization, to generate backtracking match data:

[0168] The diffusion field matrix is ​​split into window segments to obtain window-segment diffusion sub-matrices corresponding to each window segment. Then, the explanatory scale value corresponding to each window segment is mapped to the suppression coefficient of that window segment. The suppression coefficient is calculated as follows: the window suppression base value is obtained by subtracting the explanatory scale value from one; the window suppression base value is added to one to form the suppression denominator; and finally, the window suppression coefficient is obtained by dividing the window suppression base value by the suppression denominator. Thus, the larger the explanatory scale value, the smaller the window suppression coefficient, and the smaller the explanatory scale value, the larger the window suppression coefficient.

[0169] The diffusion value at each position in the diffusion submatrix of each window segment is multiplied by the window suppression coefficient of the corresponding window segment to generate the suppression diffusion matrix of that window segment. The total amount within the window is then normalized. The specific calculation process is as follows: First, all values ​​in the suppression diffusion matrix within the same window segment are summed to obtain the total amount within the window segment. Then, the total amount within the window is added to one to form the normalized denominator. Subsequently, each value at each position in the suppression diffusion matrix is ​​divided by the normalized denominator to obtain the normalized diffusion value at each position of the window segment. All normalized diffusion values ​​are arranged according to their original matrix positions to form the normalized matching matrix of that window segment. Intra-window matching aggregation is performed on the normalized matching matrix of each window segment. Specifically, all normalized diffusion values ​​in the normalized matching matrix of that window segment are summed to obtain the window matching value of that window segment. Then, the window matching values ​​of all windows are arranged according to the window segment order to generate backtracking matching data.

[0170] Perform proportion mapping modulation and point-by-point reliable modulation on the backtracking matching data and mutation proportion to generate matching reliable data. The specific calculation formula is as follows:

[0171] ;

[0172] In the formula, Indicates arch position The backtracking matching data, Indicates arch position The percentage of mutations; all the above data have been normalized during the calculation.

[0173] The reciprocal suppression mapping and window-by-window product suppression are fused between the matched reliable data and the explained proportions to generate a suppressed matching matrix. The suppressed matching matrix is ​​then aggregated to generate the source arch contribution sequence.

[0174] Perform inverse suppression mapping on the matched confidence data to generate the first suppression matrix. The specific processing method is as follows: add one to the matching confidence value corresponding to each candidate source arch in the matched confidence data for each window segment, and then take the reciprocal to obtain the first suppression value corresponding to that position. That is, the larger the matching confidence value, the smaller the first suppression value. Arrange all the first suppression values ​​according to candidate source arch × window segment to form the first suppression sequence.

[0175] The reciprocal suppression mapping process is performed on the explanatory ratio value to generate the second suppression sequence. The specific processing method is as follows: add one to the explanatory ratio value corresponding to each window segment and take the reciprocal to obtain the second suppression value corresponding to that window segment. That is, the larger the explanatory ratio value, the smaller the second suppression value. The second suppression values ​​of each window segment are arranged in the order of the window segments to form the second suppression sequence.

[0176] The second suppression sequence and the first suppression matrix are subjected to window-by-window product suppression fusion according to their positions to generate the suppression matching matrix. The calculation process is as follows: for any candidate source arch at any window segment position, the first suppression value at that position is multiplied by the second suppression value of the corresponding window segment, and the product is used as the suppression matching value of the candidate source arch in that window segment. Finally, all suppression matching values ​​are arranged in the order of candidate source arch × window segment to obtain the suppression matching matrix.

[0177] The suppression matching matrix is ​​converged to generate a source arch contribution sequence. The specific processing method is as follows: the suppression matching values ​​of each candidate source arch are summed along the window segment direction to obtain the cumulative suppression value of the candidate source arch. Then, the number of windows corresponding to the candidate source arch is counted, and the cumulative suppression value of the window segment is divided by the number of windows to obtain the average suppression contribution value of the candidate source arch. Finally, the average suppression contribution values ​​corresponding to all candidate source arches are arranged in the order of the candidate source arches to form the source arch contribution sequence.

[0178] The source arch contribution sequence is subjected to an advantage difference discrete metric to generate discrete metric values. The source arch contribution sequence and discrete metric values ​​are then encapsulated into fields to generate trend analysis results.

[0179] Sort the contribution values ​​in the source arch contribution sequence in descending order to generate a dominant sorting sequence. Then, perform a difference operation on the adjacent contribution values ​​in the dominant sorting sequence by subtracting the next contribution value from the previous one to generate a dominant difference sequence. Sum all the difference values ​​in the dominant difference sequence and divide the sum by the number of difference values ​​to generate a difference equilibrium value. Finally, subtract each difference value in the dominant difference sequence from the difference equilibrium value and take the absolute value to generate a deviation sequence.

[0180] The summation of all deviation values ​​in the deviation sequence is then divided by the number of deviation values ​​to generate a discrete metric. Field encapsulation processing is performed on the source arch contribution sequence and the discrete metric, specifically: the contribution sequence field and the discrete metric field are set according to a unified data organization order, and the source arch contribution sequence is written into the contribution sequence field and the discrete metric is written into the discrete metric field, thereby generating the trend analysis results.

[0181] This scheme generates a diffusion kernel matrix by fusing edge weight matrix data and arch position mileage data, and obtains a diffusion field matrix by diffusing the anomalous energy sequence. This enables propagation along the mutually verified edge under distance constraints while maintaining peak values. The diffusion field matrix is ​​combined with the interpretation ratio value for backtracking matching to generate backtracking matching data. This reduces the matching bias and highlights the source arch morphology when regional covariance is strong. The backtracking matching data is modulated according to the mutation ratio to generate reliable matching data. This increases the weight and suppresses noise jumps when mutation contributions are high. The reliable matching data and the interpretation ratio value are fused to generate a source arch contribution sequence and encapsulate it with discrete metrics. This enables stable locking of the source arch, diffusion direction, and influence range under sparse sensing to form trend analysis results.

[0182] The above describes the process of performing on-graph diffusion backtracking localization and encapsulation on anomalous energy sequences, edge weight matrix data, and arch mileage data to generate trend analysis results. The following describes the process of performing propagation constraint fusion on edge weight matrix data and arch mileage data to generate a diffusion kernel matrix, and then performing diffusion operations on the diffusion kernel matrix and anomalous energy sequences to generate a diffusion field matrix. Specifically, this includes:

[0183] The edge weight matrix data and the arch mileage data are fused by reciprocal decay to generate a diffusion kernel matrix, and the abnormal energy sequence and the diffusion kernel matrix are fused to generate a diffusion energy matrix.

[0184] Enhanced energy-driven peak-preserving fusion and splicing are performed on the diffusion energy matrix to generate a diffusion field matrix.

[0185] The diffusion energy matrix refers to a two-dimensional energy representation data object used to characterize the redistribution of anomalous energy in an arch group structure under graph propagation constraints, representing a spatial-temporal energy representation.

[0186] This part has already been described in detail above, so I will not repeat it here.

[0187] This scheme generates a diffusion kernel matrix by fusing edge weight matrix data and arch mileage data with reciprocal attenuation. Spatial distance is embedded into the propagation weights in a continuously attenuating manner, so that distant edges naturally weaken during diffusion while nearby edges remain dominant. This suppresses cross-regional false propagation and improves the consistency between the diffusion path and the actual adjacency relationship of the arch group. The scheme also fuses the anomalous energy sequence with the diffusion kernel matrix to generate a diffusion energy matrix. Local anomalous energy is distributed in the neighborhood according to the propagation kernel in the graph, so that the anomalous signal is transformed into a comparable spatial field. This improves the sensitivity of source arch positioning to neighborhood consistency and reduces the interference of isolated noise on positioning. The diffusion energy matrix is ​​subjected to enhanced energy-driven peak shape preservation fusion and splicing to generate a diffusion field matrix. Enhanced energy is used to perform shape-preserving weighting on the peak values ​​of key windows to avoid diffusion averaging and smoothing out peak values ​​and boundaries. This allows the diffusion field to have both propagation continuity and peak distinguishability, thereby improving the stability and positioning resolution of subsequent backtracking matching.

[0188] The above describes the propagation constraint fusion of edge weight matrix data and arch mileage data to generate a diffusion kernel matrix. Diffusion operations are then performed on the diffusion kernel matrix and the anomalous energy sequence to generate a diffusion field matrix. The following describes the reverse backtracking matching of the diffusion field matrix and the interpreted proportion values ​​to generate backtracking matching data. Finally, reliable matching modulation is performed on the backtracking matching data and the mutation proportion to generate reliable matching data. Specifically, this includes:

[0189] Perform a backtracking match on the diffusion field matrix and the explained scale value, including window suppression mapping and in-window total normalization, to generate backtracking match data;

[0190] Perform proportion mapping modulation and point-by-point reliable modulation on the backtracking matching data and mutation proportion to generate matching reliable data.

[0191] This part has already been described in detail above, so I will not repeat it here.

[0192] This scheme generates backtracking matching data by performing backtracking matching on the diffusion field matrix and interpretation ratio, which includes window segment suppression mapping and in-window total amount normalization. Window segments with strong regional covariance are weighted and suppressed using interpretation ratios, and in-window total amount normalization eliminates energy scale differences between different window segments. This makes the backtracking matching data more concentrated in reflecting the consistency of diffusion patterns formed by the propagation of anomalous source arches, thereby reducing the interference of regional convergence on source localization and improving matching comparability and stability. By performing proportion mapping modulation and point-by-point reliable modulation on the backtracking matching data and mutation proportion, reliable matching data is generated. Higher reliability weights are assigned to key windows dominated by inflection point consensus using mutation proportion, and reliability differentiation modulation is implemented for each matching point. This enables the reliable matching data to distinguish between stable propagation matching and isolated noise matching, thereby improving the noise resistance, interpretability, and source arch determination reliability of the backtracking results under sparse sensing conditions.

[0193] The above describes performing reverse backtracking matching on the diffusion field matrix and the explained proportions to generate backtracking matching data, and then performing matched confidence modulation on the backtracking matching data and the mutation proportions to generate matched confidence data. The following describes performing regional covariance suppression fusion on the matched confidence data and the explained proportions to generate source arch contribution sequences, and then performing discrete metric and encapsulation processing on the source arch contribution sequences to generate trend analysis results, specifically including:

[0194] The reciprocal suppression mapping and window-by-window product suppression are fused to match the reliable data and the explanation ratio to generate a suppression matching matrix. The suppression matching matrix is ​​then aggregated to generate the source arch contribution sequence.

[0195] The source arch contribution sequence is subjected to an advantage difference discrete metric to generate discrete metric values. The source arch contribution sequence and discrete metric values ​​are then encapsulated into fields to generate trend analysis results.

[0196] Among them, the suppression matching matrix refers to a two-dimensional data object used to suppress the regional covariance of the matching confidence of candidate source arches in each window segment;

[0197] Discrete measure refers to a data object used to characterize the degree of dispersion of the advantage of the best candidate source arch relative to the other candidate source arches in the source arch contribution sequence.

[0198] This part has already been described in detail above, so I will not repeat it here.

[0199] This scheme generates a suppressed matching matrix by fusing reciprocal suppression mapping and window-by-window product suppression with the matched reliable data and explanation ratio values. Window segments with strong regional covariance are automatically weighted using reciprocal suppression, ensuring that backtracking matching is primarily dominated by windows with more significant local differences and higher reliability. This reduces source arch misjudgments caused by overall convergence or synchronous changes in operating conditions, and improves the consistency and comparability of cross-window matching. The suppressed matching matrix is ​​then aggregated to generate a source arch contribution sequence, compressing the suppressed matching information from multiple windows into a single contribution expression, thus accumulating evidence of candidate source arches over time. The data is presented stably, reducing the impact of occasional noise in a single window segment on the ranking and improving the stability of the source arch ranking. The advantage difference discrete measurement is performed on the source arch contribution sequence to generate discrete metric values, quantifying the distinguishability between the highest contributing source arch and the other source arches, intuitively reflecting the advantage magnitude of the positioning conclusion, thereby suppressing the uncertain judgment caused by multiple source proximity and providing an auditable credibility scale. The source arch contribution sequence and discrete metric values ​​are encapsulated into fields to generate trend analysis results. The main positioning conclusion and the distinguishability evidence are presented synchronously in the same output object to ensure consistent, traceable and easy-to-verify results.

[0200] Example 2:

[0201] Please see Figure 5 A trend analysis system for long-term monitoring data of lightweight steel arches, including:

[0202] The data acquisition module is used to acquire the steel arch strain sequence and arch position mileage data of the target object;

[0203] The trend feature aggregation module is used to perform arch position mapping and segmentation processing on the steel arch strain sequence based on the arch position mileage data, generate arch strain segments, and perform symmetrical aggregation on the arch strain segments to generate a trend matrix.

[0204] The common mode trend analysis module is used to perform dual-gated edge weight reconstruction of the arch mileage data and trend matrix based on structural similarity and spatiotemporal distance, generate edge weight matrix data, and perform neighborhood consistency convergence on the trend matrix based on the edge weight matrix data to generate a common mode trend sequence.

[0205] The inflection point identification module is used to perform differential separation processing on the common mode trend sequence and the trend matrix to generate the differential mode trend sequence, and to perform on-graph inflection point consensus screening on the differential mode trend sequence and the arch strain segment to generate a consensus inflection point set.

[0206] The abnormal energy analysis module is used to perform dual-channel energy construction of differential deviation and inflection point consensus on differential trend sequence and consensus inflection point set to generate abnormal energy sequence;

[0207] The trend analysis results output module is used to perform on-map diffusion backtracking and encapsulation on abnormal energy sequences, edge weight matrix data, and arch mileage data to generate trend analysis results.

[0208] This embodiment has the same technical effects as Embodiment 1.

[0209] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The data mentioned in this application have undergone normalization and other preprocessing to unify dimensions during formula calculations.

[0210] 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 trend analysis method for long-term monitoring data of lightweight steel arches, applied to the trend analysis of continuously deployed lightweight steel arch groups in underground salt mine transportation tunnels, characterized in that... Includes the following steps: Obtain the strain sequence and arch position mileage data of the target object's steel arch; Based on the arch position mileage data, the steel arch strain sequence is subjected to arch position mapping and segmentation processing to generate arch strain segments, and symmetrical convergence is performed on the arch strain segments to generate a trend matrix. The arch mileage data and the trend matrix are subjected to dual-gated edge weight reconstruction based on structural similarity and spatiotemporal distance to generate edge weight matrix data. Then, the trend matrix is ​​subjected to neighborhood consistency convergence based on the edge weight matrix data to generate a common mode trend sequence. Perform differential separation processing on the common mode trend sequence and the trend matrix to generate a differential mode trend sequence. Perform on-graph inflection point consensus screening on the differential mode trend sequence and the arch strain segment to generate a consensus inflection point set. A dual-channel energy construction of differential deviation and inflection point consensus is performed on the differential trend sequence and the consensus inflection point set to generate an abnormal energy sequence; The abnormal energy sequence, the edge weight matrix data, and the arch mileage data are subjected to on-map diffusion backtracking positioning and encapsulation to generate trend analysis results.

2. The method for trend analysis of long-term monitoring data of lightweight steel arches according to claim 1, characterized in that: Performing symmetrical convergence on the strain segments of the arch to generate a trend matrix specifically includes: The arch strain segment is uniformly segmented to generate a window index table. Based on the window index table, the arch strain segment is separated into front and back halves and rearranged in reverse order to generate a symmetrical pairing sequence. The symmetrical paired sequences are subjected to pairwise difference and difference mean aggregation to generate a symmetrical difference set. The symmetrical difference set is then arranged in two dimensions according to the arch index and window segment index to generate a trend matrix.

3. The method for trend analysis of long-term monitoring data of lightweight steel arches according to claim 2, characterized in that: Performing a dual-gated edge weight reconstruction based on structural similarity and spatiotemporal distance on the arch mileage data and the trend matrix to generate edge weight matrix data specifically includes: The pairwise difference calculation and reciprocal decay mapping are performed on the arch mileage data to generate distance-gated data; Under the constraints of the window index table, the trend matrix is ​​extracted and rearranged within the same window, and then combined with the symmetric difference set to perform similarity mapping to generate a similarity matrix; The distance-gated data and the similarity matrix are subjected to mutual referencing gating fusion, and the mutual referencing gating fusion result is multiplied and fused with the similarity matrix to generate edge weight matrix data.

4. The method for trend analysis of long-term monitoring data of lightweight steel arches according to claim 3, characterized in that: The differential mode trend sequence and the arch strain segment are subjected to consensus screening of inflection points on the graph to generate a consensus inflection point set, specifically including: The differential trend sequence is paired with adjacent windows and the differential magnitude is measured according to the window segment index table to generate a mutation degree measure; The difference between the mean values ​​of the front and rear half windows of the arch strain segment is measured according to the window segment index table to generate a shape offset. The shape offset is then mapped to the abrupt change level to generate shape matching data. The mutation degree, the similarity matrix, and the morphological matching data are weighted and fused to generate consensus strength data. The consensus strength data is then subjected to maximum transition truncation filtering to generate a consensus inflection point set.

5. The method for trend analysis of long-term monitoring data of lightweight steel arches according to claim 4, characterized in that: Performing dual-channel energy construction of differential deviation and inflection point consensus on the differential trend sequence and the consensus inflection point set to generate anomaly energy sequences specifically includes: A unified scaling mapping is performed on the differential trend sequence and the differential normalization quantity to generate comparable deviation data; The differential mode normalization factor is obtained by performing a proximity measurement on the common mode trend sequence and the trend matrix to generate an explanatory ratio value, and then performing normalization fusion on the explanatory ratio value and the differential mode trend sequence. The gating fusion results, the consensus inflection point set, and the morphological matching data are subjected to gating consistency reinforcement mapping to generate consistent reinforcement data. The consistent reinforcement data and the comparable deviation data are then cross-linked to synthesize an abnormal energy sequence.

6. The method for trend analysis of long-term monitoring data of lightweight steel arches according to claim 5, characterized in that: Performing on-map diffusion backtracking and encapsulation on the abnormal energy sequence, the edge weight matrix data, and the arch mileage data to generate trend analysis results specifically includes: The edge weight matrix data and the arch mileage data are subjected to propagation constraint fusion to generate a diffusion kernel matrix, and diffusion operation is performed on the diffusion kernel matrix and the abnormal energy sequence to generate a diffusion field matrix; Perform reverse backtracking matching on the diffusion field matrix and the interpretation ratio value to generate backtracking matching data, and perform matching confidence modulation on the backtracking matching data and the mutation ratio to generate matching confidence data; The mutation percentage is obtained by performing a percentage mapping on the consistent reinforcement data and the comparable deviation data; Regional covariance suppression fusion is performed on the matched reliable data and the explained proportion value to generate the source arch contribution sequence, and discrete measurement and encapsulation processing is performed on the source arch contribution sequence to generate trend analysis results.

7. The method for trend analysis of long-term monitoring data of lightweight steel arches according to claim 6, characterized in that: Performing propagation constraint fusion on the edge weight matrix data and the arch mileage data to generate a diffusion kernel matrix, and performing diffusion operations on the diffusion kernel matrix and the anomalous energy sequence to generate a diffusion field matrix specifically includes: The edge weight matrix data and the arch mileage data are fused by reciprocal decay to generate a diffusion kernel matrix, and the abnormal energy sequence and the diffusion kernel matrix are fused to generate a diffusion energy matrix; Enhanced energy-driven peak-preserving fusion and splicing are performed on the diffusion energy matrix to generate a diffusion field matrix.

8. The method for trend analysis of long-term monitoring data of lightweight steel arches according to claim 6, characterized in that: Performing reverse backtracking matching on the diffusion field matrix and the interpretation ratio value to generate backtracking matching data, and performing matching confidence modulation on the backtracking matching data and the mutation ratio to generate matching confidence data specifically includes: Perform a backtracking match on the diffusion field matrix and the interpretation scale value, including window suppression mapping and in-window total normalization, to generate backtracking match data; Perform proportion mapping modulation and point-by-point reliable modulation on the backtracking matching data and the mutation proportion to generate matching reliable data.

9. The method for trend analysis of long-term monitoring data of lightweight steel arches according to claim 6, characterized in that: Performing regional covariance suppression fusion on the matched reliable data and the explained proportion values ​​to generate a source arch contribution sequence, and performing discrete metric and encapsulation processing on the source arch contribution sequence to generate trend analysis results specifically include: The reciprocal suppression mapping and window-by-window product suppression fusion are performed on the matched reliable data and the explanation ratio value to generate a suppression matching matrix, and the suppression matching matrix is ​​aggregated to generate a source arch contribution sequence; The source arch contribution sequence is subjected to an advantage difference discrete metric to generate discrete metric values. The source arch contribution sequence and the discrete metric values ​​are then encapsulated into fields to generate trend analysis results.

10. A trend analysis system for long-term monitoring data of lightweight steel arches, characterized in that, include: The data acquisition module is used to acquire the steel arch strain sequence and arch position mileage data of the target object; The trend feature aggregation module is used to perform arch position mapping and segmentation processing on the steel arch strain sequence based on the arch position mileage data, generate arch strain segments, and perform symmetrical aggregation on the arch strain segments to generate a trend matrix. The common mode trend analysis module is used to perform dual-gated edge weight reconstruction of the arch mileage data and the trend matrix based on structural similarity and spatiotemporal distance, generate edge weight matrix data, and perform neighborhood consistency convergence on the trend matrix based on the edge weight matrix data to generate a common mode trend sequence. The inflection point identification module is used to perform differential separation processing on the common mode trend sequence and the trend matrix to generate a differential mode trend sequence, and to perform on-graph inflection point consensus screening on the differential mode trend sequence and the arch strain segment to generate a consensus inflection point set. An abnormal energy analysis module is used to perform dual-channel energy construction of differential mode deviation and inflection point consensus on the differential mode trend sequence and the consensus inflection point set to generate an abnormal energy sequence. The trend analysis result output module is used to perform on-map diffusion backtracking positioning and encapsulation on the abnormal energy sequence, the edge weight matrix data, and the arch mileage data to generate trend analysis results.

Images