Dam surface deformation trends clustering analysis method based on insar monitoring data

Abstract

Disclosed in the present invention is a dam surface deformation trends clustering analysis method based on InSAR monitoring data. The method comprises: S1, establishing a monitoring system, wherein the monitoring system comprises a data acquisition module, a database module, a data-driven zoning module, and a diagnostic analysis module; S2, the data-driven zoning module analyzing variation features of monitoring data from measurement points on the basis of the received monitoring data, and performing clustering analysis on dam surface deformation trends on the basis of the variation features, and dividing a dam surface into zones having different deformation trends; and S3, the diagnostic analysis module calculating cloud parameter values of measurement points in the zones, determining whether there is a significant difference in the distribution density of cloud parameters of the same zone, and marking outliers as anomalous measurement points. With respect to the characteristics of massive time-series InSAR monitoring data, the present invention provides a targeted and efficient diagnostic analysis method by taking InSAR monitoring data of dam surface deformation as an example, so as to comprehensively and rapidly grasp dam deformation trends, and the method has adaptability and practical value for the analysis of similar multi-measurement-point monitoring data.

Description

A Cluster Analysis Method for Dam Surface Deformation Status Based on InSAR Monitoring Data Technical Field

[0001] This invention belongs to the field of water conservancy project safety monitoring technology, specifically involving a cluster analysis method for dam surface deformation patterns based on InSAR monitoring data. Background Technology

[0002] Surface deformation is the most direct reflection of structural changes in a dam, containing important evolutionary information during its service life. Scientific diagnosis and analysis of dam surface deformation monitoring data plays a crucial supporting role in understanding the dam's operational behavior, controlling engineering design parameters, and guiding the safe operation of the project.

[0003] Due to limitations in observation conditions, construction and maintenance costs, the number of traditional ground-based measuring instruments is limited, and they cannot fully reflect the overall deformation of the dam surface. Satellite remote sensing InSAR monitoring technology, as an advanced method of Earth observation, can obtain a large number of permanent scatterers or dispersed scattering targets, providing convenient conditions for full-coverage deformation monitoring of the dam surface and reflecting the true deformation situation. It has been increasingly applied to dam deformation monitoring.

[0004] Conventional monitoring data diagnostic analysis is based on single-point analysis and does not reflect the spatial correlation of the measured values. Some overall analysis methods are effective for analyzing the spatial relationship and diagnosing anomalies of a small number of ground measuring points, but on the one hand, there are differences in material properties, loads and boundary conditions in different parts of the dam body, and the resulting regional characteristics of surface deformation have not been taken into account; on the other hand, traditional ground measurement has fewer measuring points, and the adaptability of the corresponding methods to the analysis of a large number of measuring points is difficult to guarantee.

[0005] InSAR technology has a wide monitoring coverage, and the number of scattering targets extracted is often tens or even hundreds of times greater than that of ground measuring points, which is beneficial for comprehensively reflecting the deformation status of dams. However, such massive monitoring data also brings new challenges to analysis and diagnosis. Traditional methods suffer from low computational efficiency and large noise interference in long-term data, and cannot meet the needs of massive InSAR data analysis. Therefore, it is necessary to adopt a method that balances efficiency and accuracy to cluster and partition the massive InSAR monitoring data, and then further identify abnormal deformations within each partition, in order to comprehensively understand the deformation behavior of dams and provide more reliable health diagnosis results. Summary of the Invention

[0006] The purpose of this invention is to provide a cluster analysis method for dam surface deformation patterns based on InSAR monitoring data, which solves the technical problems of low accuracy and poor reliability in existing dam health status diagnosis methods.

[0007] To solve the above technical problems, the present invention adopts the following solution: A clustering analysis method for dam surface deformation based on InSAR monitoring data, comprising the following steps: Step S1: Building a monitoring system: The monitoring system includes a data acquisition module, a database module, a data partitioning module, and a diagnostic analysis module; the data acquisition module downloads SAR satellite images, extracts deformation data information of each measuring point on the dam surface using InSAR technology, and transmits the acquired monitoring data to the database module, the data partitioning module, and the diagnostic analysis module; the database module is used to store the monitoring data of the data acquisition module; Step S2: The data partitioning module analyzes the change characteristics of the monitoring data of each measuring point according to the received monitoring data, and clusters and partitions the dam surface deformation according to the change characteristics, classifying measuring points with similar change patterns into the same region, and dividing the dam surface into multiple regions with different deformation characteristics; Step S3: The diagnostic analysis module calculates the cloud parameter values ​​of the measuring points in each region, and then determines whether there are significant differences in the distribution density of the cloud parameter space in the same region, and marks outliers as abnormal measuring points.

[0008] In further optimization, in step S1, each measuring point corresponds to a unique number, and the monitoring data collected by the data acquisition module includes the deformation of each measuring point on the dam surface and the position coordinate information of each measuring point.

[0009] Further optimization, step S2 specifically includes the following steps: Step S21: Set the satellite remote sensing to extract the displacement of N measuring points on the dam surface, with a time series length of T. The displacement refers to the deformation of each measuring point relative to the initial monitoring time; then the deformation sequence of each measuring point is represented as X. i ={x i (1),x i (2),…x i (T)}, i = 1, 2, 3, ..., N.

[0010] Step S22: Construct N classes G i Each class contains a deformation sequence of one measurement point; calculate the pairwise Euclidean distance {d} between the N measurement point deformation sequences. ij}, forming the distance matrix D={d ij} N×N The specific calculation is shown in equation (1): d ij =d(G i G j )=||(X i -X ij )||2 (1); where i,j=1,2,3,…,N, and i≠j.

[0011] Step S23: Merge the two classes G with the smallest distance. e Gf This forms a new class G. e∪f , e, f = 1, 2, 3, ..., N, and e ≠ f.

[0012] Step S24: Traverse all current classes, use equation (2) to calculate the inter-class distance between the new class and any other class, and update the distance matrix; Among them, G e∪f For the new class formed in the previous step, |G e∪f | represents the number of deformable sequences contained in the new class; G e G f For the two classes used for merging, |G e |、|G f |respectively G e G f The number of deformable sequences included; G h Dividing G from the current distance matrix e G f Any other class besides |G h |For G h Number of deformable sequences included; d 2 (G e G h ) is G e G h The square of the inter-class distance, d 2 (G f G h ) is G f G h The square of the inter-class distance, d 2 (G e G f ) is G e G f The square of the inter-class distance.

[0013] Step S25: Repeat steps S23 and S24 to merge the two classes with the smallest distance in the current distance matrix to form a new class, and calculate the inter-class distance between the new class and any other class until all classes have been merged to form a cluster analysis dendrogram.

[0014] In the first round of merging, the distance d(G) between the two classes is calculated according to formula (1). e G h The first round of merging refers to merging the classes corresponding to the two sequences in step S22. The distance d(G) between the new class generated after merging and other classes is... e G h ), calculated according to formula (2).

[0015] Step S26: Based on the specific engineering conditions, trim the cluster analysis dendrogram and project the coverage of each category onto the dam surface to obtain the measurement point clustering partitioning results. The trimming should adhere to the following principles: ① Measurement point deformation sequences within the same partition are similar; ② Deformation sequences between different partitions are different; ③ The dam surface corresponding to each partition should be connected as a whole, without isolated areas, and exhibit a roughly symmetrical distribution characteristic, conforming to the deformation patterns of the dam surface.

[0016] Further optimization, step S3 specifically includes the following steps: Step S31: For any region in the clustering partition of all measurement points in step S26, assuming there are M measurement points in the partition, the deformation sequence of any measurement point a can be represented as X a ={x a (1),x a (2),…,x a (T)}, a=1,2,3,…,M; The cloud parameters of all measuring points in this area are calculated using the cloud model, and the specific calculations are shown in equations (3) to (6): Among them, E a Let a be the sample expectation of measurement point a. Let E be the sample variance of measurement point a. na Let H be the sample entropy of measurement point a. ea Let be the sample hyperentropy of measurement point a.

[0017] Step S32: Perform Z-Score standardization on the sample expectation, sample entropy, and sample hyperentropy of all measurement points in the region to obtain the standardized sample expectation E′. a Sample entropy E′ na and sample hyperentropy H′ ea The specific calculations are shown in equations (7) to (9): in, Step S33: Construct a three-dimensional cloud vector Y for each measuring point within this area. a =(E′) a E′ na H′ ea ), calculate the pairwise Euclidean distance dist(Y) between the cloud vectors of M measurement points in this region. a Y q Specifically, as shown in equation (10): dist(Y) a Y q )=||(Y a -Y q )||2 (10); where a,q=1,2,3,…,M, and a≠q.

[0018] Step S34: Convert any measurement point cloud vector Y aThe Euclidean distances to the cloud vectors of all other measurement points within the region are arranged in ascending order, where Y a The k-th Euclidean distance is k_dist(Y a ); Take all values ​​that are related to Y a The Euclidean distance does not exceed k_dist(Y) a The corresponding measurement point cloud vectors of ) form a set, which is used as Y. a k-distance neighborhood N k (Y a ), specifically as shown in formula (11): N k (Y a )={Y q |dist(Y a Y q )≤k_dist(Y a ), q≠a} (11).

[0019] Step S35: Calculate Y a Local outlier LOF(Y) a Specifically, as shown in equation (12): Among them, Y a To Y q The k-th reachable distance reach_dist k (Y q Y a ) = max(dist(Y q Y a ), k_dist(Y q )); Y a Locally achievable density Y q It is Y a k-distance neighborhood N k (Y a Any measurement point cloud vector in ); Y q Locally achievable density Similarly, Y p It is Y q k-distance neighborhood N k (Y p Any measurement point cloud vector in ); |N k (Y a )| represents N k (Y a Number of cloud vectors in |N k (Y q )| represents N k (Y p The number of cloud vectors in the cloud.

[0020] Step S36: Repeat the above steps to calculate the local outlier factor of the cloud vector of all measuring points in the region. The value represents the ratio of the mean of the local reachable density of the neighboring measuring points to that measuring point. When the local outlier factor value is greater than 1, and the larger the value, the lower the density at that measuring point is compared with the density of the neighboring measuring points. The measuring points with larger local outlier factor values ​​in the region are the more likely to be abnormal measuring points in the region and should be further identified.

[0021] Chinese patent, titled "Multi-point Joint Diagnostic System and Method for Sluice Gate Deformation", application number: 2024101723800, is the applicant's earlier application. This application is a further development based on the earlier application's solution, and there are significant differences between the two.

[0022] As a man-made structure under strict control during both construction and operation, the deformation of a dam differs significantly from that of a natural soil and rock mass like a landslide. While dam deformation inherently exhibits overall synergy, significant differences in boundary and load conditions lead to variations in deformation at different locations. For instance, at the dam crest and toe, the constraints are weaker, the boundary conditions are simpler, resulting in greater fluctuations in deformation at measuring points, with relatively synchronized deformation fluctuations between points. Conversely, at the dam toe, the constraints are stronger, the boundary conditions are more complex, and while the fluctuations in deformation at measuring points are suppressed, there is a varying degree of delay in the response to load changes, resulting in a more complex deformation process curve. Therefore, the surface deformation of a dam exhibits both overall synergy and regional characteristics.

[0023] Based on this understanding, this application first uses a clustering algorithm to divide the dam surface into several regions. Within each region, the deformation of measuring points is similar, while the deformation of measuring points differs between regions. This utilizes the regional characteristics of deformation. Within each region, there is an overall synergy in deformation among measuring points. A local anomaly factor algorithm is introduced to calculate the degree to which each measuring point differs from the deformation patterns of other measuring points. The measuring point whose deformation pattern differs most significantly from that of most measuring points is identified, and this measuring point is considered to deviate the farthest from the overall pattern, thus having the highest probability of being an anomalous measuring point.

[0024] The prior application shares the same starting point as this application, but the final results differ for two reasons. First, the prior application does not quantify the degree to which the measuring points deviate from the overall pattern; instead, it sets a fixed threshold. Points exceeding the threshold are considered abnormal. If most areas within a region are abnormal, the region is considered abnormal; otherwise, it is considered normal. Therefore, the prior application is only applicable to situations where abnormal deformation occurs in a single area. In contrast, the technical solution of this application is more flexible and adaptable. As long as the measuring point with the highest probability of abnormality is identified, whether it is abnormal is left to professionals to determine. Second, although the prior application uses clustering to divide several regions with similar deformation patterns, it does not adequately apply the regional characteristics of deformation. As mentioned earlier, the deformation fluctuations of different areas on the dam surface vary. Therefore, in the solution described in this application, whether a measuring point is normal or not is determined by comparing it with other measuring points. In contrast, the prior application judges whether the magnitude of the measuring point's fluctuation exceeds a threshold. Therefore, the appropriateness and accuracy of the threshold value will affect the precision of the final result.

[0025] Therefore, the previously applied scheme is only applicable to projects like sluice gates, which are highly regional and have a small number of measuring points, while the scheme in this application is applicable to projects like dams, which have complex deformations and numerous measuring points.

[0026] Compared with the prior art, the present invention has the following beneficial effects: 1. The monitoring system of the present invention has a high degree of intelligence, and at the same time considers the spatial correlation and regional characteristics of dam deformation. The abnormal measuring point identification effect is accurate, which solves the problems of low efficiency and unstable performance in different parts of the dam by traditional methods, and can more comprehensively reflect the deformation behavior of the dam.

[0027] 2. The method described in this invention can divide the dam body into regions with different characteristics based on the similarity of deformation at various measuring points, fully considering the regional characteristics of dam body deformation. Furthermore, the abnormal deformation diagnosis method based on cloud models and local anomaly factors in this invention has high computational efficiency and is less susceptible to noise interference, meeting the needs of InSAR data analysis and improving the accuracy and efficiency of abnormal deformation identification.

[0028] 3. This invention proposes a targeted and efficient diagnostic analysis method for massive InSAR monitoring data on dam surface deformation, balancing efficiency and accuracy, identifying abnormal deformation in each zone, comprehensively understanding the deformation behavior of the dam, and also having adaptability and practical value for similar multi-point monitoring data analysis. Attached Figure Description

[0029] Figure 1 is a schematic diagram of the dam surface deformation monitoring system based on InSAR of the present invention; Figure 2 is a distribution map of dam measuring points in Example 1; Figure 3 is a flowchart of measuring point clustering analysis; Figure 4 is a tree diagram of clustering partitions of all measuring points of the dam; Figure 5 is a clustering partition diagram of all measuring points of the dam; Figure 6 is a comparison diagram of deformation process lines of some measuring points in region IV and region V in 5 partitions; Figure 7 is a comparison diagram of deformation process lines of some measuring points in region I and region II in 6 partitions; Figure 8 is the result of cropping the tree diagram of all measuring points clustered into 5 regions; Figure 9 is a flowchart of abnormal measuring point diagnosis; Figure 10 is a histogram of LOF values ​​of measuring points in region III; Figure 11 is a schematic diagram of the location of abnormal measuring points in region III; Figure 12 is a deformation process line of some measuring points in region III. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] A clustering analysis method for dam surface deformation patterns based on InSAR monitoring data includes the following steps: Step S1: Building a monitoring system: As shown in Figure 1, the monitoring system includes a data acquisition module, a database module, a data partitioning module, and a diagnostic analysis module. The data acquisition module is used to collect SAR images of the dam location and calculate the deformation data of each measuring point on the dam surface. Each measuring point corresponds to a unique number. The monitoring data collected by the data acquisition module includes the deformation amount of each measuring point on the dam surface and the location coordinate information of each measuring point. The database module is used to store the monitoring data from the data acquisition module. The data partitioning module is used to group measuring points with similar deformation patterns into the same set, dividing the dam surface into multiple regions with different deformation characteristics. The diagnostic analysis module is used to calculate the cloud parameter values ​​of measuring points in each region, and then determine whether there are significant differences in the distribution density of cloud parameter space in the same region, and mark outliers as abnormal measuring points.

[0032] In this embodiment, a large reservoir is used as an example. The dam is a concrete-faced rockfill dam with a maximum height of 122.5m. The data acquisition module collected displacement data from 160 measuring points on the dam crest and downstream side, as shown in Figure 2. The monitoring period was from October 9, 2017 to December 30, 2022, with a total of 151 sets of data. Each set of monitoring data included deformation data from 160 measuring points. The coordinates of the measuring points were stored in the database module after the deformation data was collected.

[0033] Step S2: The data partitioning module analyzes the change characteristics of the monitoring data at each measuring point based on the received monitoring data, and performs clustering and partitioning of the dam surface deformation according to the change characteristics. Measuring points with similar change patterns are grouped into the same area, and the dam surface is divided into multiple areas with different deformation characteristics.

[0034] As shown in Figure 3, the specific steps include: Step S21: Satellite remote sensing extracted the displacement of 160 measuring points on the dam surface. The monitoring period was from October 9, 2017 to December 30, 2022. The deformation sequence of each measuring point included 151 data points, i.e., the sequence length was 151. The displacement refers to the deformation of each measuring point relative to the initial monitoring time; therefore, the deformation sequence of each measuring point is represented as X. i ={x i (1),x i (2),…x i (151)}, i = 1, 2, 3, ..., 160.

[0035] Step S22: Construct 160 classes G i Each class contains a deformation sequence of one measurement point; calculate the pairwise Euclidean distance {d} between the 160 measurement point deformation sequences. ij}, forming the distance matrix D={d ij} N×N The specific calculation is shown in equation (1). Due to space limitations, the partial data of the dam measuring point deformation sequence distance matrix is ​​shown in Table 1.

[0036] d ij =d(G i G j )=||(X i -X j )||2 (1);

[0037] Where i,j=1,2,3,…,N, and i≠j.

[0038] Table 1. Partial data of the distance matrix of the deformation sequence of dam measuring points. Step S23: Merge the two classes with the smallest distance into a new class. At this time, the Euclidean distance between the deformation sequences of measurement points numbered 56 and 71 in the matrix is ​​the smallest, which is 4.09. Therefore, measurement points 56 and 71 are merged into a new class.

[0039] Step S24: Traverse all current classes, use equation (2) to calculate the inter-class distance between the new class and any other class, and update the distance matrix;

[0040] Among them, G e∪f For the new class formed in the previous step, |G e∪f| represents the number of deformable sequences contained in the new class; G e G f For the two classes used for merging, |G e |、|G f |respectively G e G f The number of deformable sequences included; G h Dividing G from the current distance matrix e G f Any other class besides |G h |For G h Number of deformable sequences included; d 2 (G e G h ) is G e G h The square of the inter-class distance, d 2 (G f G h ) is G f G h The square of the inter-class distance, d 2 (G e G f ) is G e G f The square of the inter-class distance.

[0041] Step S25: Repeat steps S23 and S24 to merge the two classes with the smallest distance in the current distance matrix to form a new class, and calculate the inter-class distance between the new class and any other class until all classes are merged to form a cluster analysis dendrogram, as shown in Figure 4.

[0042] In the first round of merging, the distance d(G) between the two classes is calculated according to formula (1). e G h The first round of merging refers to merging the classes corresponding to the two sequences in step S22. The distance d(G) between the new class generated after merging and other classes is... e G h ), calculated according to formula (2).

[0043] Due to space limitations, only part of the inter-class distance matrix data after the first update is given, as shown in Table 2. Class 161 corresponds to the new class merged in the previous step.

[0044] Table 2. Partial data from the first update of the inter-class distance matrix of the dam measuring point deformation sequence.

[0045] Step S26: Based on the specific engineering conditions, reasonably trim the cluster analysis dendrogram and project the coverage of each category onto the dam surface to obtain the measurement point clustering partitioning results, as shown in Figure 5. The trimming should follow these principles: ① Measurement point deformation sequences in the same partition are similar; ② Deformation sequences in different partitions are different; ③ The dam surface corresponding to each partition should be connected as a whole, without isolated areas, and exhibit a roughly symmetrical distribution characteristic, conforming to the deformation law of the dam surface.

[0046] In Figure 5, different colors represent different partitions. In (a), all measuring points on the dam surface are divided into 4 regions, in (b), all measuring points on the dam surface are divided into 5 regions, and in (c), all measuring points on the dam surface are divided into 6 regions. Finally, the corresponding measuring point clustering partition results are obtained.

[0047] Comparing Figure 5(a) and (b), the 4th and 5th zones are basically the same. The difference is that regions IV and V in the 5th zone are merged in the 4th zone. However, it can be seen from the zoning map that region IV is located on the top of the dam and region V is located at the junction of the top of the dam and the left bank. The two do not have strict symmetry.

[0048] Figure 6 shows a comparison of deformation process lines of some measuring points in regions IV and V of the 5-zone system. Figure 6(a) shows the deformation process line of some measuring points in region IV; (b) shows the deformation process line of some measuring points in region V. It can be seen that the deformation trends of regions IV and V are basically consistent, exhibiting a certain periodicity. However, the measured values ​​of each measuring point in region IV are closer, and the synchronization of fluctuations is higher, but the degree of fluctuation is much higher in region IV than in region V. Based on the zoning map, it can be determined that the reason for this phenomenon is that regions IV and V are located in different places, and their boundary conditions and loads are also different. Region IV is not connected to the right bank, so the boundary conditions are simpler, and the degree of synchronization of deformation of each measuring point in the region is higher. Region V, on the other hand, is adjacent to the left bank, so it is subject to stronger constraints, and the fluctuation of deformation is smaller. In summary, the deformation trends of regions IV and V are different, and the degree of fluctuation in region IV shows a tendency to intensify. Therefore, they should not be merged into one zone for unified diagnosis.

[0049] Comparing Figure 5(b) and (c), partition 5 and partition 6 are basically the same. The difference is that regions I and II in partition 6 are merged into region I in partition 5. Observing the partition 6 diagram, it can be seen that region I is located in the inner circle and region II is located in the outer circle.

[0050] Figure 7 shows the deformation process lines of measuring points in regions I and II of the six partitions. In Figure 7, (a) is the deformation process line of measuring points in region I; and (b) is the deformation process line of measuring points in region II. It can be seen that the deformation trends of regions I and II are almost identical, with the overall deformation increasing steadily over time, while also exhibiting a certain periodicity. The only difference is that the deformation fluctuation in region II is slightly higher, but there is no trend of intensification. Furthermore, as shown in Figure 5(c), region II is divided into two disconnected regions. This partitioning result does not have good interpretability; therefore, regions I and II are suitable to be merged into one partition.

[0051] In summary, the dam measuring points should be divided into five regions, as shown in Figure 8, and the dam measuring point zoning diagram is shown in Figure 5(b). Regions I to V have 43, 35, 32, 22 and 28 measuring points, respectively.

[0052] Step S3: The diagnostic analysis module calculates the cloud parameter values ​​of each region's measuring points, and then determines whether there are significant differences in the distribution density of cloud parameters in the same region, and marks outliers as abnormal measuring points.

[0053] In this embodiment, taking 32 measuring points in region III as an example, as shown in Figure 9, the specific diagnostic analysis method includes the following steps: Step S31: For the 32 measuring points in region III, the deformation sequence of any measuring point can be represented as X a ={x a (1),x a (2),…,x a (T)}, a=1,2,3,…,32; The cloud parameters of all measuring points in this area are calculated using the cloud model, and the specific calculations are shown in equations (3) to (6):

[0054] Among them, E a Let x be the sample expectation of measurement point a. a (b) represents the deformation value of measuring point a at time b, S a 2 Let E be the sample variance of measurement point a. na Let H be the sample entropy of measurement point a. ea The sample hyperentropy is for measurement point a. The cloud parameter calculation results for all measurement points in Region III are shown in Table 3.

[0055] Table 3. Calculation results of cloud parameters for all measuring points in Region III.

[0056] Step S32: Perform Z-Score standardization on the sample expectation, sample entropy, and sample hyperentropy of all measurement points in the region to obtain the standardized sample expectation E′. a Sample entropy E′ naand sample hyperentropy H′ ea The specific calculations are shown in equations (7) to (9): in, Step S33: Construct a three-dimensional cloud vector Y for each measuring point within this area. a =(E′) a E′ na H′ ea ), calculate the pairwise Euclidean distance dist(Y) between the cloud vectors of the 32 measuring points in this region. a Y q Specifically, as shown in equation (10): dist(Y) a Y q )=||(Y a -Y q )||2 (10); where a,q=1,2,3,…,32, and a≠q.

[0057] The cloud vector distance calculation results for some measuring points in Region III are shown in Table 4.

[0058] Table 4. Vector Distance Matrix Data of Measurement Points in Region III

[0059] Step S34: Convert any measurement point cloud vector Y a The Euclidean distances to the cloud vectors of all other measurement points within the region are arranged in ascending order, where Y a The k-th Euclidean distance is k_dist(Y a ); Take all values ​​that are related to Y a The Euclidean distance does not exceed k_dist(Y) a The corresponding measurement point cloud vectors of ) form a set, which is used as Y. a k-distance neighborhood N k (Y a Here, k is set to 4, as shown in equation (11): N k (Y a )={Y q |dist(Y a Y q )≤k_dist(Y a ), q≠a} (11).

[0060] Step S35: Calculate Y a Local outlier LOF(Y) a Specifically, as shown in equation (12): Among them, Y a To Y q The k-th reachable distance reach_distk (Y q Y a ) = max(dist(Y q Y a ), k_dist(Y q )); Y a Locally achievable density Y q It is Y a k-distance neighborhood N k (Y a Any measurement point cloud vector in ); Y q Locally achievable density Similarly, Y p It is Y q k-distance neighborhood N k (Y p Any measurement point cloud vector in ); |N k (Y a )| represents N k (Y a Number of cloud vectors in |N k (Y q )| represents N k (Y p The number of cloud vectors in the cloud.

[0061] The LOF values ​​of all measurement point cloud vectors in Region III are shown in Table 5.

[0062] Table 5. LOF values ​​of cloud vectors from measuring points at the dam in Region III.

[0063] Step S36: Calculate the Local Outlier Factor (LOF) of the cloud vectors of all measurement points within the region. a The histogram of the LOF values ​​of the measuring points is shown in Figure 10. The measuring point #42, with the largest LOF value, was identified as a suspected abnormal measuring point, and its location is shown in Figure 11. The measurement process lines of some measuring points in Area III are shown in Figure 12. The deformation of measuring point #42 has been significantly higher than that of other measuring points since September 2019, and it shows a rapid increasing trend, which is significantly different from other measuring points, indicating that the identification is accurate. Further analysis can be conducted in conjunction with the engineering operation situation.

[0064] Similarly, follow the steps above to diagnose whether there are frequently tested points in other areas, which will not be repeated here.

[0065] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the technical concept of this invention. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A cluster analysis method for dam surface deformation patterns based on InSAR monitoring data, characterized in that, Includes the following steps: Step S1: Set up the monitoring system: The monitoring system includes a data acquisition module, a database module, a data partitioning module, and a diagnostic analysis module; the data acquisition module downloads SAR satellite images, uses InSAR technology to extract deformation data information of various measuring points on the dam surface, and transmits the acquired monitoring data to the database module, the data partitioning module, and the diagnostic analysis module; the database module is used to store the monitoring data from the data acquisition module; Step S2: The data partitioning module analyzes the change characteristics of the monitoring data of each measuring point based on the received monitoring data, and performs clustering and partitioning of the dam surface deformation according to the change characteristics. Measuring points with similar change patterns are divided into the same area, and the dam surface is divided into multiple areas with different deformation characteristics. Step S3: The diagnostic analysis module calculates the cloud parameter values ​​of each region's measuring points, and then determines whether there are significant differences in the distribution density of cloud parameters in the same region, and marks outliers as abnormal measuring points.

2. The method for cluster analysis of dam surface deformation patterns based on InSAR monitoring data according to claim 1, characterized in that, In step S1, each measuring point corresponds to a unique number, and the monitoring data collected by the data acquisition module includes the deformation of each measuring point on the dam surface and the position coordinate information of each measuring point.

3. The method for cluster analysis of dam surface deformation patterns based on InSAR monitoring data according to claim 2, characterized in that, Step S2 specifically includes the following steps: Step S21: Set the satellite remote sensing to extract displacement data from N measuring points on the dam surface, with a time series length of T. The displacement refers to the deformation of each measuring point relative to the initial monitoring time. Then, the deformation sequence of each measuring point is represented as X. i ={x i (1),x i (2),…x i (T)}, i = 1, 2, 3, ..., N; Step S22: Construct N classes G i Each class contains a deformation sequence of one measurement point; calculate the pairwise Euclidean distance {d} between the N measurement point deformation sequences. ij }, forming the distance matrix D={d ij } N×N The specific calculation is shown in equation (1): d ij =d(G i ,G j )=||(X i -X j )||2 (1); Where i,j=1,2,3,…,N, and i≠j; Step S23: Merge the two classes G with the smallest distance. e G f This forms a new class G. e∪f e,f = 1,2,3,…,N, and e≠f; Step S24: Traverse all current classes, use equation (2) to calculate the inter-class distance between the new class and any other class, and update the distance matrix; Among them, G e∪f For the new class formed in the previous step, |G e∪f | represents the number of deformable sequences contained in the new class; G e G f For the two classes used for merging, |G e |、|G f |respectively G e G f The number of deformable sequences included; G h Dividing G from the current distance matrix e G f Any other class besides |G h |For G h Number of deformable sequences included; d 2 (G e G h ) is G e G h The square of the inter-class distance, d 2 (G f G h ) is G f G h The square of the inter-class distance, d 2 (G e G f ) is G e G f The square of the inter-class distance; Step S25: Repeat steps S23 and S24 to merge the two classes with the smallest distance in the current distance matrix to form a new class, and calculate the inter-class distance between the new class and any other class until all classes have been merged to form a cluster analysis dendrogram; Step S26: Cut the cluster analysis dendrogram according to the specific engineering conditions, and project the coverage of each category onto the dam surface to obtain the clustering partitioning results of the measurement points.

4. The method for cluster analysis of dam surface deformation patterns based on InSAR monitoring data according to claim 3, characterized in that, Step S3 specifically includes the following steps: Step S31: For any region in the clustering partition of all measurement points in step S26, assuming there are M measurement points in the partition, the deformation sequence of any measurement point a can be represented as X. a ={x a (1),x a (2),…,x a (T)}, a=1,2,3,…,M; The cloud parameters of all measuring points in this area are calculated using the cloud model, and the specific calculations are shown in equations (3) to (6): Among them, E a S is the expected value of the sample at measurement point a. a 2 Let E be the sample variance of measurement point a. na Let H be the sample entropy of measurement point a. ea The sample hyperentropy at measurement point a; Step S32: Perform Z-Score standardization on the sample expectation, sample entropy, and sample hyperentropy of all measurement points in the region to obtain the standardized sample expectation E′. a Sample entropy E′ na and sample hyperentropy H′ ea The specific calculations are shown in equations (7) to (9): in, Step S33: Construct a three-dimensional cloud vector Y for each measuring point within this area. a =(E′) a E′ na H′ ea ), calculate the pairwise Euclidean distance dist(Y) between the cloud vectors of M measurement points in this region. a ,Y q Specifically, as shown in equation (10): dist(Y a ,AND q )=||(Y a -AND q )||2 (10); Where a,q=1,2,3,…,M, and a≠q; Step S34: Convert any measurement point cloud vector Y a The Euclidean distances to the cloud vectors of all other measurement points within the region are arranged in ascending order, where Y a The k-th Euclidean distance is k_dist(Y a ); Take all values ​​that are related to Y a The Euclidean distance does not exceed k_dist(Y) a The corresponding measurement point cloud vectors of ) form a set, which is used as Y. a k-distance neighborhood N k (Y a Specifically, as shown in equation (11): N k (AND a ){AND q |dist(Y a ,AND q )≤k_dist(Y a ),q≠a} (11); Step S35: Calculate Y a Local outlier LOF(Y) a Specifically, as shown in equation (12): Among them, Y a To Y q The k-th reachable distance reach_dist k (Y q ,Y a ) = max(dist(Y q ,Y a ),k_dist(Y q )); Y a Locally achievable density Y q It is Y a k-distance neighborhood N k (Y a Any measurement point cloud vector in ); Y q Locally achievable density Similarly, Y p It is Y q k-distance neighborhood N k (Y p Any measurement point cloud vector in ); |N k (Y a )| represents N k (Y a Number of cloud vectors in |N k (Y q )| represents N k (Y p Number of cloud vectors in the middle; Step S36: Repeat the above steps to calculate the local outlier factor of the cloud vector of all measuring points in the region. The value represents the ratio of the mean of the local reachable density of the neighboring measuring points to that measuring point. When the local outlier factor value is greater than 1, and the larger the value, the lower the density at that measuring point is compared with the density of the neighboring measuring points. The measuring points with larger local outlier factor values ​​in the region are the more likely to be abnormal measuring points in the region and should be further identified.

Images