Bridge group deformation mode evaluation method and system based on PS-InSAR and complex network theory

Abstract

The invention relates to the field of data mining and machine learning, and relates to a bridge group deformation mode evaluation method and system based on PS-InSAR and complex network theories. The method comprises the following steps of: 1, performing refined preprocessing on high-dimensional time series data of engineering structure monitoring points obtained by multiple sources; 2, constructing a behavior pattern distribution vector of each monomer structure by adopting a UMAP-KDPI algorithm; and step 3, constructing a group association network by calculating behavior similarity between structures, and performing community discovery, key node identification and multi-level association characteristic deep analysis by using a network science theory to realize bridge group deformation mode evaluation. The problem that in an existing monitoring data analysis method for an engineering structure group, potential and common behavior patterns in complex high-dimensional time series data are difficult to effectively recognize, and an effective means for accurately quantifying and comparing behavior similarity among structure individuals is lacked is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining and machine learning, and specifically to a bridge group deformation mode assessment method and system based on PS-InSAR and complex network theory. Background Art

[0002] Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR), a large-scale, long-time, high-precision Earth observation technology, has been widely used in urban infrastructure settlement, landslide monitoring, and geological hazard assessment. In recent years, with the development of high-resolution SAR satellite constellations and the continuous maturity of algorithms, research on the application of PS-InSAR technology in bridge structural health monitoring has deepened and established a certain research status. Currently, research on bridge monitoring using PS-InSAR technology focuses on the following areas: First, long-term deformation monitoring of single large bridges. By identifying high-density PS points at key bridge locations (such as towers, main beams, and piers), millimeter-level deformation time series are obtained to assess the bridge's overall settlement, linear deformation rate, and thermodynamic expansion and contraction behavior. Second, the overall stability of multiple bridges within a specific area is surveyed. Through large-scale PS-InSAR processing, bridges with abnormal deformation rates within the area are identified, providing maintenance departments with preliminary clues for risk investigation. Third, PS-InSAR monitoring results are cross-validated with traditional monitoring methods such as GPS and leveling to enhance the reliability of monitoring data. These studies have fully demonstrated the effectiveness and potential of PS-InSAR technology as a macroscopic, long-term bridge deformation monitoring tool.

[0003] However, an in-depth analysis of the current state of PS-InSAR bridge monitoring research reveals significant shortcomings. These limitations limit the full potential of PS-InSAR technology for the intelligent operation and maintenance of urban bridge clusters and systemic risk assessment. First, the vast majority of existing studies remain at the "point" or "island" level of analysis, lacking a perspective and methodology for systematically assessing the bridge cluster as a whole. Existing studies often treat each bridge as an independent analysis object, evaluating its own indicators such as deformation rate and cumulative deformation, or horizontally comparing these indicators across different bridges. This analytical approach ignores the inherent interconnectedness of urban bridge clusters as complex infrastructure networks. For example, bridges located in the same geological conditions or subject to the same regional groundwater fluctuations may exhibit similar settlement patterns; bridges with the same design type or construction process may experience similar disease evolution pathways under similar loads or environmental conditions. Existing analytical methods struggle to effectively identify, quantify, and model these behavioral interconnections among bridges, failing to reveal the collective behavioral characteristics, coordinated response mechanisms, and potential systemic risks of bridge clusters at the "network" and "system" levels.

[0004] Secondly, existing research is seriously lacking in its ability to identify the complex deformation patterns contained in PS-InSAR monitoring points. Currently, most studies on the analysis of PS-InSAR time series data focus on fitting and extracting linear deformation rates. Although this method can reflect long-term macroscopic deformation trends, it ignores a large amount of more valuable periodic and stage-by-stage deformation information. These refined deformation patterns are the key basis for judging the current health status of bridges, identifying early damage, and warning of future risks. However, automatically, accurately, and unsupervisedly mining and classifying these diverse deformation patterns from massive high-dimensional time series data covering tens of thousands of PS points is a huge challenge and technical gap in the current PS-InSAR data processing and analysis process. Existing analysis methods often rely on tedious manual interpretation or simple statistical thresholds, which are inefficient and unreliable.

[0005] Third, due to the lack of effective identification of detailed deformation patterns and quantification of inter-bridge behavioral correlations, existing research has been unable to construct assessment models that reflect the true interactions among bridge groups. This results in regional bridge maintenance decisions often being ranked based solely on apparent metrics (such as maximum settlement rate) for individual bridges, without considering the importance of the bridge within the overall network, the specificity of its behavior, or its interconnectedness with other bridges. This decision-making approach can lead to uneven resource allocation, neglecting bridges that, despite experiencing minimal deformation, are located in critical network locations or exhibit early signs of collective anomalies, thus compromising the resilience and safety of the entire urban transportation network. Summary of the Invention

[0006] The present invention provides a bridge group deformation pattern assessment method and system based on PS-InSAR and complex network theory, which are used to solve the problems in existing monitoring data analysis methods for engineering structure groups, such as the difficulty in effectively identifying potential and common behavioral patterns in complex high-dimensional time series data, the lack of effective means to accurately quantify and compare the behavioral similarities between individual structures, and the difficulty in revealing the complex subgroup structure, key nodes and systematic correlation characteristics within the structure group.

[0007] The present invention is achieved through the following technical solutions: A bridge group deformation mode assessment method based on PS-InSAR and complex network theory, the method comprising the following steps: Step 1: Perform fine preprocessing on the high-dimensional time series data of engineering structure monitoring points acquired from multiple sources; Step 2: Use the UMAP-KDPI algorithm to construct the behavioral pattern distribution vector of each monomer structure; Step 3: Construct a group association network by calculating the behavioral similarity between structures, and apply network science theory to perform community discovery, key node identification, and in-depth analysis of multi-level association characteristics to achieve deformation pattern evaluation of bridge groups.

[0008] Furthermore, step 1 specifically includes collecting monitoring data from multiple engineering structures, where the monitoring data includes structural response values ​​of multiple monitoring points on each structure at multiple time stamps, and may include spatial coordinate information of the monitoring points; cleaning, aligning and filtering the original data to extract valid monitoring point time series data and coordinate information.

[0009] Furthermore, the step 2 includes the following steps: Step 2.1: UMAP manifold structure reconstruction and embedding. Specifically, the standard UMAP algorithm achieves dimensionality reduction by constructing a weighted graph and optimizing low-dimensional embedding. Step 2.2: Co-optimization of clustering objectives and manifold learning, specifically, feeding back or integrating the KMeans clustering objective into the UMAP low-dimensional embedding optimization process; Step 2.3: KMeans clustering in the embedding space and adaptive determination of the number of clusters k. Specifically, KMeans clustering is performed in the low-dimensional embedding space obtained after the UMAP-KDPI co-optimization. Step 2.4: Standardization processing, specifically, before executing the UMAP-KDPI algorithm, the time series data of all input monitoring points are standardized; Step 2.5: After global clustering processing, refine and extract the distribution characteristics of the multi-dimensional behavioral patterns of the monomer structure.

[0010] Furthermore, the step 2.1 is specifically as follows: Step 2.1.1: Distance metric selection. Specifically, considering the characteristics of time series deformation data, in addition to Euclidean distance, we can explore the use of dynamic time warping distance or its variants as the initial distance metric when constructing high-dimensional graphs using UMAP. Step 2.1.2: Neighborhood and connectivity maintenance. Specifically, UMAP defines the connection strength between points in high-dimensional space through fuzzy set theory. UMAP-KDPI maintains the original UMAP balance between local and global structures while adjusting the n_neighbors and min_dist parameters or introducing additional regularization terms to enhance the density and separation of potential "cluster core" areas, preparing for subsequent embedded clustering.

[0011] Furthermore, the step 2.2 is specifically as follows: Step 2.2.1: Iterative optimization. Specifically, after several rounds of UMAP iteratively constructing low-dimensional embeddings, tentatively run KMeans on the current embedding space and use the KMeans clustering results as an indicator of embedding quality. Step 2.2.2: Joint loss function. Specifically, the joint loss function includes both the topology preservation loss term of UMAP itself and the clustering loss term of KMeans.

[0012] Furthermore, the step 2.3 is specifically as follows: Step 2.3.1: Determine the number of clusters k. Specifically, the embedding properties of UMAP-KDPI allow for a more intuitive or adaptive identification of the "natural" number of clusters in low-dimensional space by observing the distribution density and separation of data points, or by using density-based methods. Alternatively, the algorithm may have built-in exploration of the optimal solution to the joint loss function under different values ​​of k during the collaborative optimization process, thereby recommending an optimal value of k. Step 2.3.2: Clustering is performed. Specifically, the KMeans algorithm is used to perform the final cluster assignment on the optimized low-dimensional embedding points to obtain the global behavior pattern cluster label to which the time series data of each monitoring point belongs.

[0013] Furthermore, the step 3 specifically includes the following steps: Step 3.1: Based on the extracted behavioral pattern distribution vectors of each structure, a similarity measurement method is used to calculate the behavioral similarity between any two structures, where the similarity measurement method includes cosine similarity or similarity based on Jensen-Shannon divergence; Step 3.2: Calculate the behavioral similarity between any two structures through step 1, and construct and analyze the group association network; Step 3.3: Based on the constructed group association network, realize analysis visualization and result output.

[0014] A bridge group deformation mode assessment system based on PS-InSAR and complex network theory, the system using the above-mentioned bridge group deformation mode assessment method based on PS-InSAR and complex network theory, the system comprising: Preprocessing module: performs refined preprocessing on high-dimensional time series data of engineering structure monitoring points acquired from multiple sources; Monomer structure behavior pattern distribution vector module: uses the UMAP-KDPI algorithm to construct the behavior pattern distribution vector of each monomer structure; Bridge group deformation pattern assessment module: This module constructs a group association network by calculating the behavioral similarity between structures, and applies network science theory to perform community discovery, key node identification, and in-depth analysis of multi-level association characteristics to achieve bridge group deformation pattern assessment.

[0015] A deformation pattern of a bridge group evaluated using the above method is applied to the safety monitoring data analysis of urban bridge groups.

[0016] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above method is implemented.

[0017] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0018] The beneficial effects of the present invention are: This invention is suitable for the long-term monitoring data analysis of various large and extra-large engineering structure groups. It aims to achieve intelligent recognition of their complex temporal behavior patterns, accurate measurement of behavioral similarities between individuals, and in-depth analysis and visualization of group association networks, providing strong support for the intelligent operation and maintenance and risk assessment of regional infrastructure.

[0019] The bridge group deformation pattern assessment technology based on PS-InSAR and complex network theory described in the present invention adopts a unified heterogeneous data integration and analysis framework, and integrates the standardized preprocessing and feature extraction module of multi-source and multi-type engineering structure monitoring data into the front end of the overall analysis process. This module is used to perform consistency regularization and preliminary refinement of common features on the original, significantly different structural monitoring data, so that the subsequent UMAP-KDPI core algorithm can perform cross-structure behavior pattern mining and comparison based on a standardized data, thereby realizing efficient, universal analysis and unified management of large-scale, heterogeneous engineering structure group monitoring data.

[0020] The bridge group deformation pattern assessment technology based on PS-InSAR and complex network theory described in the present invention adopts the construction technology of structural-level behavioral pattern distribution vector (PMV) to integrate and quantitatively characterize the UMAP-KDPI clustering results of all monitoring points within each single structure. The PMV vector is used as a "fingerprint" to describe the overall behavioral pattern of the structure to accurately calculate the behavioral similarity between different structures (such as through cosine similarity and JS similarity), so that the behavioral differences and correlation degrees between structures can be objectively and quantifiably evaluated, achieving effective measurement and in-depth understanding of the complex behavioral correlations between groups of engineering structures.

[0021] The bridge group deformation pattern assessment technology based on PS-InSAR and complex network theory described in the present invention adopts a group association analysis model based on network science to convert the calculated inter-structural behavioral similarity matrix into a weighted engineering structure group association network. It uses community discovery algorithms (such as Louvain) and network centrality analysis techniques to deeply analyze the constructed network topology structure, so that the sub-group divisions, core key structures and potential systemic risk transmission paths within the structure group can be clearly revealed, realizing cross-level association cognition and risk insight from individual behavior to group system characteristics.

[0022] The bridge group deformation pattern assessment technology based on PS-InSAR and complex network theory described in the present invention adopts a multimodal, interactive visualization analysis engine to integrate multi-level analysis results (including pattern space distribution maps, PMV characteristics, similarity heat maps, network topology maps, UMAP embedding distribution maps, etc.) from monitoring points to single structures to structural groups into a unified visualization interface. Interactive technologies such as linkage highlighting and drill-down query are used to intuitively present and explore complex data and analysis results, allowing users to easily understand behavioral patterns, structural associations and group trends, and achieve effective communication of analysis results and data-driven insight discovery.

[0023] The bridge group deformation pattern assessment technology based on PS-InSAR and complex network theory described in the present invention adopts a modular and process-oriented algorithm design concept, standardizes and integrates key steps such as data acquisition, preprocessing, UMAP-KDPI pattern recognition, PMV construction, similarity calculation, network analysis and visualization, and uses optimized algorithm implementation and parameter adaptation mechanism (part) to efficiently schedule and execute the entire analysis process. This significantly reduces manual intervention in the analysis process of large-scale engineering structure group monitoring data, improves processing efficiency, and realizes an improvement in the automation and intelligence level of structure group behavior analysis.

[0024] The bridge group deformation pattern assessment technology based on PS-InSAR and complex network theory described in the present invention adopts the core UMAP-KDPI ensemble learning algorithm to deeply couple and collaboratively optimize the objectives of the KMeans clustering process with the improved UMAP manifold learning process. This collaborative optimization mechanism is used to perform nonlinear dimensionality reduction and embedded clustering on high-dimensional and complex monitoring point time series data, enabling the algorithm to automatically identify and strengthen the intrinsic cluster structure corresponding to deformation patterns of different physical significance in the data, thereby achieving in-depth mining and precise identification of potential, common, and subtle behavioral patterns of engineering structures that were previously difficult to discover through traditional thresholds or simple statistical methods.

[0025] The bridge group deformation pattern assessment technology based on PS-InSAR and complex network theory described in the present invention adopts an analysis result output and interpretation module oriented to actual operation and maintenance needs, organically combining identified behavioral patterns, structural health profiles, group synergy effect assessments, systemic risk assessments, and network key node information with engineering practice knowledge. These comprehensive analysis results provide scientific data support and clear action recommendations for key decision-making links such as regional infrastructure health assessment, maintenance priority sorting, abnormal structure warnings, and disaster emergency response, making engineering operation and maintenance management more forward-looking, accurate, and effective, and achieving the ultimate goal of improving infrastructure safety and resilience and optimizing resource allocation.

[0026] The bridge group deformation pattern assessment technology described in this invention, based on PS-InSAR and complex network theory, is suitable for analyzing long-term monitoring data of various large and extra-large engineering structure groups (such as urban bridge groups, building groups, dam groups, etc.). It aims to achieve intelligent identification of their complex temporal behavior patterns, accurate measurement of behavioral similarities between individuals, and in-depth analysis and visualization of group association networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of the method of the present invention.

[0028] Figure 2 This is a schematic diagram of the unified temporal behavior pattern clustering process of the present invention, showing the analysis results of data normalization and UMAP-KDPI.

[0029] Figure 3 This is an example of the spatial distribution diagram of the behavior pattern of the monomer engineering structure of the present invention.

[0030] Figure 4 Schematic diagram of the four structural behavior mode curves of the present invention.

[0031] Figure 5 This is an example of a heat map of the similarity matrix between structures calculated based on cosine similarity in the present invention.

[0032] Figure 6 This is an example of a heat map of the inter-structure similarity matrix calculated based on Jensen-Shannon similarity in the present invention.

[0033] Figure 7 is a visualization example of the engineering structure group association network under the two indicators of the present invention, and the nodes are colored according to the Louvain community. Figure 7 (a) is a visualization example of the engineering structure group association network embedded in the complex network community of the bridge group based on cosine similarity, and Figure 7 (b) is a visualization example of the engineering structure group association network embedded in the complex network community of the bridge group based on JSD similarity.

[0034] Figure 8 This is an example of visualizing the two-dimensional embedding space after UMAP dimensionality reduction of the structural behavior pattern distribution vector in the present invention. The nodes are colored according to the Louvain community.

[0035] Figure 9 This is a scatter plot comparing the results of two similarity metrics (cosine similarity and JS similarity) in the present invention. DETAILED DESCRIPTION

[0036] In the following description, specific details such as specific system structures and technologies are provided for illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present application with unnecessary details.

[0037] It will be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0038] It should also be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0039] The following is a clear and complete description of the technical solutions in the embodiments of this application in conjunction with the drawings in the specification of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0040] In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0041] Implementation Method 1 This century's approach proposes a bridge group deformation pattern assessment method based on PS-InSAR and complex network theory. Complex network theory is incorporated into the PS-InSAR bridge group data analysis framework. By constructing and analyzing a behavioral correlation network for a bridge group, this method achieves a paradigm shift from "point-based" monitoring of individual bridges to a systematic, networked assessment of bridge groups. This technology aims to address the challenges of existing PS-InSAR applications, which include the inability to effectively identify complex deformation patterns, quantify behavioral correlations between bridges, and conduct systematic risk assessment of groups. This technology transforms massive, seemingly isolated PS-InSAR monitoring point data into deep insights into the internal structure, group behavior, key nodes, and potential risk transmission pathways of the entire bridge group. Its significance lies in providing urban managers and engineers with an unprecedented macroscopic perspective and scientific tools. This allows for more accurate assessment of the overall health of regional infrastructure, identification of systemic vulnerabilities, and optimization of maintenance resource allocation strategies. It also provides data-driven decision support for resilient design of urban transportation networks and emergency management, significantly enhancing the safety and intelligent management capabilities of urban infrastructure.

[0042] Combine Figures 1 to 9 To describe this embodiment, faced with massive amounts of high-dimensional PS-InSAR time-series data, this paper employs an advanced, improved unsupervised manifold learning and embedded clustering algorithm to automatically identify complex deformation patterns in bridge clusters. This algorithm goes beyond simply combining dimensionality reduction and clustering in tandem. Instead, through a collaborative optimization mechanism, it deeply integrates clustering objectives into the manifold embedding process. This approach aims to directly mine highly discriminative and physically meaningful fundamental deformation patterns (such as stability, linear settlement, periodic fluctuations, and accelerated deformation) from the high-dimensional data, and assigns a pattern classification label to each PS monitoring point.

[0043] Secondly, after accurately identifying and classifying the deformation patterns of all monitoring points, this paper proposes a method for quantitatively defining the behavioral characteristics of individual bridges. By calculating the proportion or composition of different deformation patterns across all PS points on each bridge, a unique "pattern distribution vector," or "behavioral fingerprint," is generated for each bridge, comprehensively reflecting its overall behavioral characteristics.

[0044] Finally, the key step in this invention lies in constructing a complex network model of the bridge cluster. Each bridge is treated as a node in the network. Based on the "behavioral fingerprint" generated in the previous step, the similarity (e.g., cosine similarity) between any two bridges' behavioral fingerprints is calculated to quantify the strength of their behavioral associations. When this association strength exceeds a preset threshold, a connecting edge is established between the corresponding bridge nodes, with the weight representing the association strength. In this way, the abstract behavioral relationships of the entire bridge cluster are transformed into a concrete, analyzable complex network.

[0045] Finally, once the complex network of bridge clusters is constructed, the present invention utilizes established network science analysis tools to conduct in-depth exploration of the network. For example, community discovery algorithms (such as Louvain) are used to identify subgroups (communities) of bridges with highly similar behavioral patterns, and the distribution patterns of these communities across geographic space and structural types are analyzed. Centrality metrics (such as degree centrality and betweenness centrality) are calculated for each node to identify core bridges that play a pivotal role in the bridge cluster network. These analytical results ultimately constitute a systematic, network-based assessment of the deformation patterns of the bridge cluster, thereby achieving the core objective of the present invention.

[0046] This embodiment provides a bridge group deformation mode assessment method based on PS-InSAR and complex network theory, the method comprising the following steps: Step 1: Perform fine preprocessing on the high-dimensional time series data of engineering structure monitoring points acquired from multiple sources; Step 2: Use the UMAP-KDPI algorithm (an improved combination of manifold learning and embedded clustering) to intelligently identify and cluster the unified temporal behavior patterns of monitoring points, and based on this, construct the behavior pattern distribution vector of each monomer structure; Step 3: Construct a group association network by calculating the behavioral similarity between structures, and apply network science theory to perform community discovery, key node identification, and in-depth analysis of multi-level association characteristics to achieve deformation pattern evaluation of bridge groups.

[0047] Furthermore, step 1 specifically includes collecting monitoring data from multiple engineering structures (such as a group of bridges), where the monitoring data includes structural response values ​​(such as deformation, displacement, etc.) of multiple monitoring points on each structure at multiple time stamps, and may include spatial coordinate information of the monitoring points (such as geometric data in WKT format); cleaning, aligning, and filtering the original data to extract valid monitoring point time series data and coordinate information.

[0048] Furthermore, the step 2 includes the following steps: Step 2.1: UMAP manifold structure reconstruction and embedding. Specifically, the standard UMAP algorithm achieves dimensionality reduction by constructing a weighted graph and optimizing low-dimensional embedding. Step 2.2: Co-optimization of clustering objectives and manifold learning, specifically, feeding back or integrating the KMeans clustering objective into the UMAP low-dimensional embedding optimization process; Step 2.3: KMeans clustering in the embedding space and adaptive determination of the number of clusters k. Specifically, KMeans clustering is performed in the low-dimensional embedding space obtained after the UMAP-KDPI co-optimization. Step 2.4: Standardization processing, specifically, before executing the UMAP-KDPI algorithm, the time series data of all input monitoring points are standardized; Standardize the time series data of all monitoring points of all structures to eliminate dimensional differences; For example, using Z-score normalization:

[0049] Where X is the original time series data, is the mean, is the standard deviation; Step 2.5: After global clustering processing, refine and extract the distribution characteristics of the multi-dimensional behavioral patterns of the monomer structure.

[0050] Furthermore, the step 3 specifically includes the following steps: Step 3.1: Based on the extracted behavioral pattern distribution vectors of each structure, calculate the behavioral similarity between any two structures using a similarity measurement method, where the similarity measurement method includes cosine similarity or similarity based on Jensen-Shannon divergence (JS Similarity); Step 3.2: Calculate the behavioral similarity between any two structures through step 1, and construct and analyze the group association network; Step 3.3: Based on the constructed group association network, realize analysis visualization and result output.

[0051] In this implementation, the core of the method is to use the original UMAP-KDPI algorithm to perform unified and in-depth behavioral pattern recognition on the high-dimensional time series data of monitoring points of large-scale engineering structure groups (such as urban bridge groups), and combine structural level feature extraction, similarity measurement, network science analysis and interactive visualization technology to achieve systematic evaluation and intelligent decision support from single-point behavior to group association.

[0052] In this implementation, the behavioral pattern refers to the time series characteristics of deformation, displacement, strain, and other responses exhibited by monitoring points over a certain time span. Examples include continuous settlement, periodic reciprocating motion, accelerated deformation, stable micro-motion, and other types of dynamic behavior. The resulting assessment results can be used to support risk assessment, maintenance priority sorting, disaster warning, and operation and maintenance management optimization for regional infrastructure.

[0053] Combine Figures 1 to 9 Further explanation, the step 2.1 is specifically as follows: Step 2.1.1: Distance metric selection. Specifically, considering the characteristics of time series deformation data, in addition to Euclidean distance, we can explore the use of dynamic time warping distance or its variants as the initial distance metric when constructing high-dimensional graphs using UMAP. Step 2.1.2: Neighborhood and connectivity maintenance. Specifically, UMAP defines the connection strength between points in high-dimensional space through fuzzy set theory. UMAP-KDPI maintains the original UMAP balance between local and global structures while adjusting the n_neighbors and min_dist parameters or introducing additional regularization terms to enhance the density and separation of potential "cluster core" areas, preparing for subsequent embedded clustering.

[0054] Furthermore, the step 2.2 is specifically as follows: Step 2.2.1: Iterative optimization. Specifically, after several rounds of UMAP iteratively constructing low-dimensional embeddings, tentatively run KMeans on the current embedding space and use the KMeans clustering results as an indicator of embedding quality. Step 2.2.2: Joint loss function. Specifically, the joint loss function includes both the topology preservation loss term of UMAP itself and the clustering loss term of KMeans.

[0055] Furthermore, the step 2.3 is specifically as follows: Step 2.3.1: Determine the number of clusters k. Specifically, the embedding properties of UMAP-KDPI allow for a more intuitive or adaptive identification of the "natural" number of clusters in low-dimensional space by observing the distribution density and separation of data points, or by using density-based methods. Alternatively, the algorithm may have built-in exploration of the optimal solution to the joint loss function under different values ​​of k during the collaborative optimization process, thereby recommending an optimal value of k. Step 2.3.2: Clustering is performed. Specifically, the KMeans algorithm is used to perform the final cluster assignment on the optimized low-dimensional embedding points to obtain the global behavior pattern cluster label to which the time series data of each monitoring point belongs.

[0056] Furthermore, the step 3.2 is specifically as follows: Step 3.2.1: Network construction: Each engineering structure is regarded as a node in the network, and the behavioral similarity between structures is used as the weight of the connecting edge; when the similarity is greater than the preset threshold When , a connection edge is established between the corresponding nodes; Step 3.2.2: Community Discovery: Use community discovery algorithms, such as the Louvain algorithm, to divide the constructed structural association network into communities and identify structural subgroups with similar behavioral pattern clustering characteristics; Step 3.2.3: Network topology characteristics analysis; Calculate the centrality index of each node in the network (such as degree centrality, weighted degree centrality, eigenvector centrality, betweenness centrality, etc.) to evaluate the importance and influence of each structure in the group. For example, the weighted degree centrality of a node Defined as:

[0057] in, is the set of neighbor nodes of node i, is the weight of the edge connecting node i and node j.

[0058] Furthermore, the step 3.3 is specifically as follows: Step 3.3.1: Spatial distribution map of monomeric structural patterns: For monitoring points with spatial coordinate information, visualize their spatial positions on the structure and mark their global cluster labels with different colors to form a spatial distribution map of the monomeric structural behavior patterns; Step 3.3.2: Similarity matrix heat map: Visualize the calculated structural similarity matrix in the form of a heat map; Step 3.3.3: Visualize the association network: Visualize the association network of the constructed engineering structure groups. You can color the nodes according to the community division results and adjust the edge thickness or color according to the similarity. Step 3.3.4: Visualization of embedded behavioral patterns of structural groups: The obtained structural-level pattern distribution vector can be further reduced in dimension (such as UMAP) and visualized in two-dimensional or three-dimensional space to observe the overall distribution trend and aggregation of structural groups, and can be colored according to the community division results.

[0059] More specifically, Furthermore, a preferred embodiment is provided, wherein the multi-source monitoring data acquisition and refined pre-processing of step 1 specifically includes: Step 1.1 Multi-data compatible access: By designing a compatible data interface, we can achieve access and unified formatting of engineering structure monitoring data from multiple sources, including Excel / CSV (including WKT geometry information), Shapefile, and monitoring system database. The data interface is designed to be compatible with multiple common structural monitoring data formats, including but not limited to: a) Excel (.xlsx, .xls) or CSV (.csv) spreadsheet files containing WKT (Well-Known Text) formatted geometry columns and time series data columns identified by a specific prefix (such as "D" or "Deform_"); b) Shapefile (.shp) GIS file containing monitoring point geometry objects and attribute tables storing time series data; c) Obtain data directly from the structural health monitoring system database (e.g., SQL database, time series database) via API or query. For data from different sources, perform a unified internal data structure conversion to form a standardized dataset containing the monitoring point ID, the structure ID to which it belongs, spatial coordinates (X, Y, Z, if applicable), timestamp, and corresponding monitoring values ​​(e.g., deformation, strain, acceleration, etc.).

[0060] Step 1.2: Perform quality improvement operations on the acquired raw time series data, including identifying and processing outliers (such as elimination or interpolation) using statistical or time series decomposition methods, filling missing data with linear interpolation, spline interpolation, or model-based predictive interpolation techniques, and completing benchmark alignment and unified frequency resampling of the time series at each monitoring point. Time series data quality improvement and alignment: Strict quality control of the raw time series data, including: a) Outlier detection and processing: Statistical methods (such as the 3σ principle and the IQR rule) or time series decomposition methods (such as STL decomposition) are used to identify and process abnormal spikes or drifting data caused by sensor failures, transmission errors, etc. The processing methods can be elimination, smoothing, or interpolation based on adjacent data. b) Missing data filling: For missing data segments caused by data collection or transmission interruptions, appropriate interpolation algorithms (such as linear interpolation, spline interpolation, predictive interpolation based on the ARIMA model, or spatiotemporal interpolation methods for PS-InSAR data) are selected to fill the gaps based on the length of the missing segments and the data characteristics to ensure the integrity and continuity of the time series. c) Time base alignment and resampling: For asynchronous data from different monitoring systems or different monitoring points, unify the time base and resample all time series to a unified time frequency (such as daily, weekly, or monthly) through interpolation or aggregation according to analysis requirements.

[0061] Step 1.3, Spatial Information Verification and Registration: If the monitoring points contain spatial coordinate information, perform a unified coordinate system conversion (for example, from a local coordinate system to a unified geographic coordinate system such as WGS84 or the National Geodetic Coordinate System). Coordinate data obtained from different sources should be verified for consistency and, if necessary, registered to ensure the accuracy of spatial analysis. If the data source only provides relative locations or no coordinate information, the method can still be executed, but some spatial visualization functions will be limited.

[0062] Step 1.4. Screening of valid monitoring points and analysis periods: Based on the completeness of monitoring data coverage, signal-to-noise ratio, structural representativeness of monitoring points, and the specific requirements of the analysis task (e.g., before and after a specific event, in a specific season, etc.), screen a subset of valid monitoring points and core analysis periods for subsequent analysis, and eliminate redundant data of poor quality or low relevance to the analysis objectives.

[0063] Furthermore, a preferred embodiment is provided in which the unified temporal behavior pattern recognition and clustering based on UMAP-KDPI in step 2 is a key step in the core UMAP-KDPI algorithm of the present invention. The implementation details of the core UMAP-KDPI algorithm include: Step 2.1, UMAP manifold structure reconstruction and embedding: The standard UMAP algorithm achieves dimensionality reduction by constructing a weighted graph and optimizing low-dimensional embedding. In the UMAP-KDPI of the present invention, this process is improved: a) Distance metric selection: In addition to Euclidean distance, considering the characteristics of time series deformation data, we can explore the use of dynamic time warping (DTW) distance or its variants as the initial distance metric when constructing high-dimensional graphs using UMAP to better capture phase inconsistencies and morphological similarities between time series. b) Neighborhood and Connectivity Preservation: UMAP uses fuzzy set theory to define the strength of connections between points in high-dimensional space and strives to maintain this connection strength in low-dimensional space. While maintaining the original UMAP's balance between local and global structure, UMAP-KDPI may enhance the density and separation of regions considered potential "cluster cores" by adjusting the n_neighbors (number of neighbors) and min_dist (minimum distance between embedded points in low-dimensional space) parameters or introducing additional regularization terms, preparing for subsequent embedded clustering.

[0064] Step 2.2: Co-optimization of clustering objective and manifold learning: The key to UMAP-KDPI is to feed back or integrate the KMeans clustering objective (minimizing the intra-cluster sum of squares) into the UMAP low-dimensional embedding optimization process. This can be achieved through one or a combination of the following approaches: a) Iterative Optimization: After several rounds of UMAP iteratively constructing low-dimensional embeddings, KMeans is tentatively run on the current embedding space. The KMeans clustering results (such as the clarity of cluster assignments, inter-cluster distances, and intra-cluster variance) are used as a metric to evaluate the quality of the embedding. This metric is then used to adjust the UMAP loss function, thereby guiding UMAP to generate embeddings that are more conducive to KMeans distinguishing patterns. b) Joint loss function: A joint loss function is designed that combines UMAP's own topology-preserving loss term and KMeans' clustering loss term (or its proxy indicator). By optimizing this joint loss function, the dimensionality reduction and clustering processes are carried out in a coordinated manner. The resulting low-dimensional embedding not only maintains the original data structure but also naturally presents a clear cluster structure.

[0065] Step 2.3, KMeans clustering in the embedding space and adaptive determination of the number of clusters k: KMeans clustering is performed in the low-dimensional embedding space obtained after the UMAP-KDPI co-optimization. Since the embedding space has been guided and constructed with enhanced cluster separability, the following are the results: a) Determining the number of clusters k: This may no longer rely solely on traditional external evaluation metrics (such as the silhouette coefficient and Calinski-Harabasz index, which are used to evaluate multiple k values ​​through trial and error). The embedding properties of UMAP-KDPI may enable a more intuitive or adaptive identification of the "natural" number of clusters in low-dimensional space by observing the distribution density and degree of separation of data points, or by utilizing density-based methods (such as DBSCAN, which identifies high-density core regions). Alternatively, the algorithm may incorporate into the collaborative optimization process the exploration of the optimal solution to the joint loss function for different k values, thereby recommending an optimal k value.

[0066] b) Clustering execution: Use the KMeans algorithm to perform the final cluster assignment on the optimized low-dimensional embedding points to obtain the global behavior pattern cluster label to which the time series data of each monitoring point belongs.

[0067] Step 2.4, normalization: Before executing the UMAP-KDPI algorithm, Z-score normalization is performed on the time series data of all input monitoring points to eliminate the differences in dimensions and absolute values, so that the algorithm focuses more on the deformation pattern itself rather than the absolute value.

[0068] Where X is the original time series data, is the mean of the series, is the standard deviation of the series.

[0069] Step 2.5: Based on the above, refine and extract the distribution characteristics of the multi-dimensional behavioral patterns of the monomer structure.

[0070] Furthermore, a preferred embodiment is provided, wherein the step 2.5 of refining and extracting the multi-dimensional behavior pattern distribution characteristics of the monomer structure specifically includes: Step 2.5.1. Construction of the basic mode distribution vector (PMV): For each engineering structure, count the absolute number or normalized ratio of all valid monitoring points belonging to each global behavior pattern cluster (obtained from 2) to form the basic mode distribution vector of the structure. ,in represents the proportion of monitoring points belonging to the jth global cluster in structure s, and k is the total number of global clusters. This vector preliminarily characterizes the overall tendency of the structure in different identified behavior patterns.

[0071] Step 2.5.2, Spatially Weighted Pattern Distribution Vector (SPMV) Construction: If the monitoring points have spatial coordinates and importance weights (for example, based on prior knowledge such as the key position of the monitoring point in the structure and the contribution to the overall response of the structure), the cluster affiliation of each monitoring point can be weighted when calculating the pattern distribution vector. For example

[0072] in is the set of monitoring points belonging to cluster j, Indicates that monitoring point i belongs to structure s, is the spatial or importance weight of monitoring point i. This makes the pattern distribution vector better reflect the behavior pattern of the key area of ​​the structure.

[0073] Step 2.5.3, Time Evolving Pattern Distribution Feature (TPMV) Extraction: In addition to static pattern distribution, we can also analyze the changes in pattern distribution over time. For example, the entire monitoring period is divided into several sub-periods, and the behavior pattern distribution vector of the structure is calculated independently or updated in each sub-period, thereby obtaining a set of distribution vector sequences that evolve over time. This can reveal migrations or shifts in patterns of structural behavior.

[0074] Step 2.5.4, feature vector fusion and dimensionality reduction: If multiple features such as basic PMV, SPMV, TPMV are considered at the same time, the feature vector dimension may be too high. In this case, feature selection methods can be used or feature-level dimensionality reduction can be performed again (such as shallow application of PCA or UMAP) to obtain a more compact but information-rich structure-level refined behavioral feature vector. , used for subsequent similarity calculation.

[0075] Furthermore, a preferred embodiment is provided, wherein the integrated measurement of multi-index behavioral similarity between structures in step 3.1 specifically includes: Step 3.1.1. Similarity calculation based on distribution vector: Use the structure-level refined behavior feature vector obtained in step 3 , calculates the similarity between any two structures (such as structure a and structure b). The core measurement methods include: a) Cosine similarity: , measures the consistency of vector direction and is applicable to proportional data; b) and / or Jensen-Shannon (JS) similarity: First, calculate the JS divergence between the two probability distributions (make sure the vector elements are non-negative and sum to 1, which is naturally satisfied if the PMV is proportional).

[0076] in , is the Kullback-Leibler divergence. Then it is converted to similarity

[0077] in JS divergence JS similarity is more sensitive to the difference in distribution.

[0078] Step 3.1.2: Normalize and Sparsify the Similarity Matrix: The values ​​in the original similarity matrix may be distributed over varying ranges. Normalization can be performed (e.g., scaling to the interval [0, 1]). To facilitate network analysis and visualization, a threshold can be set based on the similarity score, treating similarities below the threshold as zero. This allows for sparsification of the similarity matrix, retaining only the most significant structural associations.

[0079] Furthermore, a preferred embodiment is provided, wherein the multi-level association network construction and in-depth analysis of the engineering structure group in step 3.2 specifically includes: Step 3.2.1. Weighted undirected / directed network construction: Using the engineering structure as the network node, use the (integrated) similarity score calculated by 4 As the weight of the edge connecting structures a and b (e.g., ANALYSIS_EDGE_THRESHOLD). If the similarity is symmetric (e.g., cosine similarity, JS similarity), an undirected (or directed) network is constructed. If asymmetric associations are considered (e.g., the behavioral pattern of structure A significantly influences structure B, but not vice versa, which may require more complex causal inference or time evolution analysis), a directed network can be constructed. One or more weight thresholds can be set to filter out weak connections and form a meaningful network topology.

[0080] Step 3.2.2, Louvain algorithm maximizes modularity Q To optimize community division:

[0081] in, are the elements of the adjacency matrix, is the degree of node i, m is the total number of edges (or total weight) in the network, is the community to which node i belongs, is an indicator function, when 1 if yes, 0 otherwise.

[0082] Step 3.2.3: Robustness Assessment of Key Node and Subgroup Identification: Perform robustness assessments on key nodes (e.g., high-centrality nodes) and identified communities (subgroups) in the network. For example, by making small perturbations to edge weights or randomly removing some nodes / edges, observe the stability of key node ranking and community structure to enhance the reliability of the analysis results.

[0083] Step 3.2.4: Network Motif Analysis and Functional Subgraph Mining: Search for frequently occurring specific connectivity patterns (network motifs, such as small three- or four-node subgraphs) within the structural association network. These motifs may correspond to specific types of interaction patterns within the structural group. Furthermore, subgraphs with specific "functions" can be mined, for example, identifying clusters of "hub structures" that bridge different communities within the network, or clusters of tight "core structures" that exhibit highly homogeneous behavior.

[0084] Step 3.3.5. Correlation analysis with external factors: Conduct statistical correlation analysis on network topological characteristics (such as node centrality and community affiliation) or community behavior pattern characteristics with known external influencing factors (such as the design age of the structure, the geological zone in which it is located, records of major surrounding engineering activities, extreme climate events, etc.) to explore the impact mechanism of these external factors on the behavioral patterns and interactions of structural groups.

[0085] Furthermore, a preferred embodiment is provided, wherein the multimodal visual analysis and intelligent decision support result output of step 3.3 specifically includes: Step 3.3.1. Interactive spatiotemporal visualization platform: Develop or integrate an interactive visualization platform that can display: a) The geographical location of each monomer structure (map view) and the spatial distribution of its monitoring point behavior pattern; b) Detailed deformation time series curve of the selected structure or monitoring point and the UMAP-KDPI behavior pattern cluster information to which it belongs c) a correlation network diagram of the entire bridge cluster, with nodes optionally colored and sized by community, centrality index, or risk level associated with external factors; d) Similarity matrix heat map, and support clicking on the heat map cell to view the detailed information of the corresponding structure pair; e) The distribution of structures (based on their PMV vectors) in the UMAP embedding space also supports interactive exploration. Users can perform multi-dimensional and multi-granular drill-down analysis from macroscopic groups to microscopic individuals through operations such as clicking, swiping, and zooming.

[0086] Step 3.3.2: Structural Health "Profiling" and Early Warning: Based on the behavioral pattern clusters identified by UMAP-KDPI, combined with the corresponding deformation characteristics of each cluster (such as average deformation rate, volatility, and trend) and engineering experience, each pattern cluster is assigned a preliminary semantic label indicating health status or risk level (such as "stable," "slow settlement," "periodic fluctuation," or "accelerated deformation risk"). Furthermore, based on the pattern distribution vectors (PMVs) of individual structures and their community affiliation and centrality in the associated network, a comprehensive assessment of their health status is conducted to form a "digital profile" of the structural health. Early warnings are issued for structures that exhibit a continuously increasing proportion of high-risk behavioral patterns or that migrate from stable communities to high-risk communities.

[0087] Step 3.3.3, Group Synergy Effect and Systemic Risk Assessment Report: Output a comprehensive analysis report that not only includes the assessment of the individual structures, but also focuses on revealing the group synergy effects and systemic risks. For example: a) Identify the main structural communities and their common behavioral characteristics, and analyze possible causes (e.g., similar designs, same geological conditions); b) A list of key influential nodes (bridges) and their contribution to network stability; c) If there is a regional subsidence or common disaster pattern, quantify its scope and extent; d) Predict potential risk transmission paths or chain reaction possibilities based on network structure.

[0088] Step 3.3.4: Prioritize Operations and Maintenance Decisions: Combine the health profiling from Step 3.3.2 and the group risk assessment from Step 3.3.3 to provide data-driven support for maintenance decisions. For example, prioritize detailed inspections and allocated maintenance resources to structures exhibiting high-risk behavior patterns, located in critical network locations, or belonging to high-risk communities. For groups of structures exhibiting similar early abnormal behavior patterns, targeted preventive measures can be implemented in batches.

[0089] The specific embodiment is: Take the PS-InSAR monitoring dataset of multiple bridges in a city (for example, 50 bridges of different types) as an example: 1. Data Preparation: Monthly deformation data from PS-InSAR monitoring points on eight bridges over the past three years were collected. The data were stored as Excel files, one file per bridge, containing the WKT coordinates of the monitoring points and columns of deformation values ​​in DDYYMM format. Using the preprocessing process described in Implementation 2, the data was cleaned, interpolated (e.g., linear interpolation was used for values ​​missing for less than three consecutive months), resampled (to monthly values), and removed from monitoring points with less than 70% of the valid time series length. This resulted in a high-quality time series dataset containing, for example, 8,000 valid monitoring points.

[0090] 2. UMAP-KDPI pattern recognition (corresponding to Figure 2 The Unified Mapping and Kernel Dependent Inference (UMAP) algorithm was applied to the standardized (Z-score) time series vectors of all 32,256 monitoring points (each vector is 35-dimensional, corresponding to 35 months). UMAP-KDPI parameter settings included: for example, using the Euclidean distance in the UMAP stage, adaptively adjusting n_neighbors based on data density, setting min_dist to 0.05, and setting the target embedding dimension to 2. The KMeans component of the collaborative optimization, leveraging the inherent mechanisms of UMAP-KDPI, identified k=4 primary clusters of deformation behavior patterns. Figure 2 The distribution of all monitoring points in the UMAP-KDPI two-dimensional embedding space can be displayed and colored according to 4 clusters.

[0091] 3. Bridge-level PMV construction (corresponding to Figure 4 ): For each bridge, calculate the proportion of all its monitoring points belonging to these five behavior pattern clusters to form a 4-dimensional PMV vector.

[0092] 4. Calculation of similarity between bridges (corresponding to Figure 5 、 Figure 6 ): Based on the PMV vectors of 8 bridges, the cosine similarity matrix is ​​calculated respectively and JS similarity matrix . Figure 5 and Figure 6 The heat maps of these two 8×8 similarity matrices are displayed separately, and the similarity can be represented by the depth of color, and can be supplemented by a hierarchical clustering tree diagram.

[0093] 5. Bridge Association Network Analysis (corresponding to Figure 7): A bridge association network was constructed for eight bridges. The Louvain algorithm was applied for community discovery, identifying two major bridge communities. Weighted degree centrality and betweenness centrality were calculated for each bridge node. Figure 7 shows the bridge network diagram, with node size corresponding to centrality metrics and color corresponding to community.

[0094] 6. Visualization and decision support (corresponding to Figure 3 、 Figure 8 、 Figure 9 ): Figure 3 Showing a specific bridge, with monitoring points colored by the four behavioral pattern clusters identified by UMAP-KDPI, showing the spatial distribution of the patterns.

[0095] Figure 8 Shows the distribution of all 8 bridges (based on their PMV vectors) in the 2D embedding space after UMAP secondary dimensionality reduction, and colors them according to the 2 network communities they belong to.

[0096] Figure 9 Shows the scatter plot comparison of cosine similarity and JS similarity on all bridge pairs.

[0097] Generate reports: Indicate which bridges exhibit high-risk deformation patterns (e.g., a cluster identified by experts as "continuously accelerated settlement"), which bridge communities exhibit abnormal overall performance, and which bridges are located at critical connection points in the network, providing maintenance departments with inspection and maintenance priority recommendations.

[0098] Implementation Method 2 This embodiment provides a bridge group deformation mode assessment system based on PS-InSAR and complex network theory. The system uses the bridge group deformation mode assessment method based on PS-InSAR and complex network theory as described in the first embodiment. The system includes: Preprocessing module: performs refined preprocessing on high-dimensional time series data of engineering structure monitoring points acquired from multiple sources; Monolithic structure behavior pattern distribution vector module: This module uses the UMAP-KDPI algorithm (an improved combination of manifold learning and embedded clustering) to intelligently identify and cluster the unified temporal behavior patterns of monitoring points, and based on this, constructs the behavior pattern distribution vectors of each monolithic structure. Bridge group deformation pattern assessment module: This module constructs a group association network by calculating the behavioral similarity between structures, and applies network science theory to perform community discovery, key node identification, and in-depth analysis of multi-level association characteristics to achieve bridge group deformation pattern assessment.

[0099] Implementation Method 3 This embodiment provides a bridge group deformation pattern evaluated by the method described in the first embodiment and applied to the safety monitoring data analysis of urban bridge groups.

[0100] Implementation Method 4 An embodiment of the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. The memory is used to store software programs and modules, and the processor executes various functional applications and data processing by executing the software programs and modules stored in the memory. The memory and processor are connected via a bus. Specifically, the processor implements any step of the first embodiment described above by executing the computer program stored in the memory.

[0101] It should be understood that in the embodiments of the present invention, the processor referred to herein may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0102] The memory may include a read-only memory, a flash memory, and a random access memory, and provides instructions and data to the processor. A portion or all of the memory may also include a non-volatile random access memory.

[0103] It should be understood that if the above-mentioned integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present invention can also implement all or part of the processes in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The above-mentioned computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The above-mentioned computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The above-mentioned computer-readable medium can include: any entity or device capable of carrying the above-mentioned computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content contained in the above-mentioned computer-readable storage medium can be appropriately increased or decreased based on the requirements of legislation and patent practice in a jurisdiction.

[0104] The above description of the disclosed embodiments will enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein, but is intended to be construed in the widest manner consistent with the principles and novel features disclosed herein.

Claims

1. A bridge group deformation mode assessment method based on PS-InSAR and complex network theory, characterized by: The method comprises the following steps: Step 1: Perform fine preprocessing on the high-dimensional time series data of engineering structure monitoring points acquired from multiple sources; Step 2: Use the UMAP-KDPI algorithm to construct the behavioral pattern distribution vector of each monomer structure; Step 3: Construct a group association network by calculating the behavioral similarity between structures, and apply network science theory to perform community discovery, key node identification, and in-depth analysis of multi-level association characteristics to achieve deformation pattern evaluation of bridge groups.

2. The method according to claim 1, characterized in that Specifically, step 1 includes collecting monitoring data from multiple engineering structures, wherein the monitoring data includes structural response values ​​of multiple monitoring points on each structure at multiple time stamps, and may include spatial coordinate information of the monitoring points; The raw data is cleaned, aligned and filtered to extract valid monitoring point time series data and coordinate information.

3. The method according to claim 1, characterized in that The step 2 comprises the following steps: Step 2.1: UMAP manifold structure reconstruction and embedding. Specifically, the standard UMAP algorithm achieves dimensionality reduction by constructing a weighted graph and optimizing low-dimensional embedding. Step 2.2: Co-optimization of clustering objectives and manifold learning, specifically, feeding back or integrating the KMeans clustering objective into the UMAP low-dimensional embedding optimization process; Step 2.3: KMeans clustering in the embedding space and adaptive determination of the number of clusters k. Specifically, KMeans clustering is performed in the low-dimensional embedding space obtained after the UMAP-KDPI co-optimization. Step 2.4: Standardization processing, specifically, before executing the UMAP-KDPI algorithm, the time series data of all input monitoring points are standardized; Step 2.5: After global clustering processing, refine and extract the distribution characteristics of the multi-dimensional behavioral patterns of the monomer structure.

4. The method according to claim 1, characterized in that The step 2.1 is specifically as follows: Step 2.1.1: Distance metric selection. Specifically, considering the characteristics of time series deformation data, in addition to Euclidean distance, we can explore the use of dynamic time warping distance or its variants as the initial distance metric when constructing high-dimensional graphs using UMAP. Step 2.1.2: Neighborhood and connectivity preservation. Specifically, UMAP defines the connection strength between points in high-dimensional space through fuzzy set theory. UMAP-KDPI maintains the original UMAP balance between local and global structures while adjusting the n_neighbors and min_dist parameters or introducing additional regularization terms to enhance the density and separation of potential "cluster core" areas, preparing for subsequent embedded clustering.

5. The method according to claim 4, characterized in that: The step 2.2 is specifically as follows: Step 2.2.1: Iterative optimization. Specifically, after several rounds of UMAP iteratively constructing low-dimensional embeddings, tentatively run KMeans on the current embedding space and use the KMeans clustering results as an indicator of embedding quality. Step 2.2.2: Joint loss function. Specifically, the joint loss function includes both the topology preservation loss term of UMAP itself and the clustering loss term of KMeans.

6. The method according to claim 4, characterized in that: The step 2.3 is specifically as follows: Step 2.3.1: Determine the number of clusters k. Specifically, the embedding properties of UMAP-KDPI allow for a more intuitive or adaptive identification of the "natural" number of clusters in low-dimensional space by observing the distribution density and separation of data points, or by using density-based methods. Alternatively, the algorithm may have built-in exploration of the optimal solution to the joint loss function under different values ​​of k during the collaborative optimization process, thereby recommending an optimal value of k. Step 2.3.2: Clustering is performed. Specifically, the KMeans algorithm is used to perform the final cluster assignment on the optimized low-dimensional embedding points to obtain the global behavior pattern cluster label to which the time series data of each monitoring point belongs.

7. The method according to claim 1, characterized in that The step 3 specifically includes the following steps Step 3.1: Based on the extracted behavioral pattern distribution vectors of each structure, a similarity measurement method is used to calculate the behavioral similarity between any two structures, where the similarity measurement method includes cosine similarity or similarity based on Jensen-Shannon divergence; Step 3.2: Calculate the behavioral similarity between any two structures through step 1, and construct and analyze the group association network; Step 3.3: Based on the constructed group association network, realize analysis visualization and result output.

8. A bridge group deformation mode assessment system based on PS-InSAR and complex network theory, characterized by: The system uses the bridge group deformation mode assessment method based on PS-InSAR and complex network theory as described in any one of claims 1 to 7, and the system includes: Preprocessing module: performs refined preprocessing on high-dimensional time series data of engineering structure monitoring points acquired from multiple sources; Monomer structure behavior pattern distribution vector module: uses the UMAP-KDPI algorithm to construct the behavior pattern distribution vector of each monomer structure; Bridge group deformation pattern assessment module: This module constructs a group association network by calculating the behavioral similarity between structures, and applies network science theory to perform community discovery, key node identification, and in-depth analysis of multi-level association characteristics to achieve bridge group deformation pattern assessment.

9. A computer device, characterized in that The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images