Method and system for reconstructing surface continuous deformation map

Abstract

The present application belongs to the technical field of geological disaster monitoring, and specifically discloses a reconstruction method and system of a surface continuous deformation map, which comprises the following steps: determining a region to be reconstructed according to a surface deformation map of a study area; training a surface subsidence prediction model based on historical surface deformation data and historical clustered multi-source heterogeneous data of the study area; inputting current clustered multi-source heterogeneous data of the region to be reconstructed into the trained subsidence prediction model to obtain a predicted deformation map of the region to be reconstructed; and fusing the surface deformation map and the predicted deformation map to obtain a surface continuous deformation map of the study area. The present application can make full use of multi-source heterogeneous environmental factor data, reduce the dependence on existing historical data, cluster multi-source heterogeneous data to greatly reduce the influence of spatial heterogeneity of different source factors on the construction of a surface subsidence prediction model, and use machine learning modeling and prediction to make the prediction results more reliable and more accurate.

Description

Technical Field

[0001] This invention belongs to the field of geological disaster monitoring technology, specifically a method for reconstructing continuous surface deformation maps based on a multi-factor random forest regression model. It can be applied to the recovery of continuous deformation fields of geological disasters over a large area, and is used for geological disaster investigation, monitoring and prevention. Background Technology

[0002] Traditional methods for monitoring land surface deformation, such as Global Navigation Satellite System (GNSS) and leveling, are limited by their low spatial coverage and resolution, making them unsuitable for large-scale continuous deformation monitoring. Interferometric Synthetic Aperture Radar (InSAR) offers significant advantages in high spatial resolution and large-scale deformation monitoring. Currently, InSAR technology is widely used in urban land subsidence monitoring. However, InSAR technology faces two major technical challenges in continuous land surface deformation monitoring: firstly, long spatiotemporal baselines, vegetation, and surface disturbances lead to decoherence; secondly, large deformation gradients cause phase unwrapping errors, resulting in missing monitoring data in InSAR land surface deformation maps. This significantly limits the spatiotemporal feature analysis and mechanism inversion research of InSAR deformation maps.

[0003] To address this problem, existing technologies typically employ spatial interpolation or machine learning methods to predict and reconstruct surface deformation maps. Firstly, conventional interpolation methods, such as inverse distance (IDW) interpolation and Kriging interpolation, rely on the similarity of statistical or geometric structures between known and unknown data for prediction, resulting in predictions with only mathematical significance. However, the prediction results of interpolation methods are easily affected by the spatial distribution of known measurement data. They perform best when the measurements are uniformly distributed; if the observation data is sparse and unevenly distributed, the accuracy of the interpolation results decreases. Furthermore, conventional machine learning surface deformation prediction models generally ignore the spatial heterogeneity of surface subsidence caused by changes in groundwater levels and the distribution of ground fissures, leading to lower accuracy in surface deformation prediction.

[0004] However, in large-scale InSAR surface deformation measurements, the distribution of coherence points is not uniform, and the influencing factors and mechanisms of surface deformation in different regions are also different, making it difficult for existing methods to obtain high-precision surface deformation reconstruction results. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for reconstructing a continuous surface deformation map, so as to solve the technical problem that existing methods are difficult to obtain high-precision surface deformation reconstruction results.

[0006] A first aspect of the present invention provides a method for reconstructing a continuous surface deformation map, comprising:

[0007] Step 1: Determine the area to be reconstructed based on the surface deformation map of the study area;

[0008] Step 2: Train the surface subsidence prediction model based on historical surface deformation data and historical clustered multi-source heterogeneous data of the study area;

[0009] Step 3: Input the current clustered multi-source heterogeneous data of the area to be reconstructed into the trained settlement prediction model to obtain the predicted deformation map of the area to be reconstructed.

[0010] Step 4: Fuse the surface deformation map and the predicted deformation map to obtain a continuous surface deformation map of the study area.

[0011] Preferably, step 1 specifically includes:

[0012] The surface deformation map of the study area is divided into multiple regions;

[0013] Areas that do not contain surface deformation data are identified as areas to be reconstructed.

[0014] Preferably, the method further includes the following step before step 2:

[0015] Obtain historical multi-source heterogeneous datasets;

[0016] From the historical multi-source heterogeneous dataset, identify multiple historical heterogeneous data from different sources that have a correlation with surface deformation data greater than a preset first threshold.

[0017] Clustering is performed on historical heterogeneous data from multiple different sources that exceed a preset first threshold to obtain clustered multi-source heterogeneous data.

[0018] Preferably, from the historical multi-source heterogeneous dataset, multiple historical heterogeneous data from different sources that have a correlation with surface deformation data greater than a preset first threshold are identified, specifically including:

[0019] Obtain the frequency ratio of each type of historical heterogeneous data in the historical multi-source heterogeneous dataset in the surface deformation map;

[0020] Based on the correlation between the frequency ratio and the surface subsidence rate value, historical heterogeneous data from multiple different sources with a correlation greater than a preset first threshold are identified.

[0021] Preferably, obtaining the frequency ratio of each type of historical heterogeneous data in the historical multi-source heterogeneous dataset in the surface deformation map specifically includes:

[0022] Obtain the first ratio between the number of settlement areas affected by each type of historical heterogeneous data in the historical multi-source heterogeneous dataset and the total number of settlement areas;

[0023] Obtain the second ratio between the number of regions occupied by each type of historical heterogeneous data in the historical multi-source heterogeneous dataset and the total number of regions in the surface deformation map;

[0024] Based on the first ratio and the second ratio, determine the frequency ratio of each type of historical heterogeneous data in the surface deformation map.

[0025] Preferably, clustering is performed on heterogeneous data from multiple different sources that exceed a preset first threshold, specifically including:

[0026] The K-Means clustering method is used to cluster heterogeneous data from multiple different sources that exceed a preset first threshold.

[0027] Preferably, before clustering historical heterogeneous data from multiple different sources that exceed a preset first threshold, the method further includes:

[0028] The correlation between any two historical heterogeneous data sets from multiple different sources is obtained when the correlation with the surface subsidence rate value is greater than a preset first threshold.

[0029] When the correlation is greater than the preset second threshold, one of the two corresponding historical heterogeneous data is selected to obtain multiple historical heterogeneous data from different sources after correlation processing.

[0030] Accordingly, clustering is performed on multiple historical heterogeneous data from different sources that exceed a preset first threshold, specifically as follows:

[0031] Clustering is performed on historical heterogeneous data from multiple different sources after correlation processing.

[0032] Preferably, the settlement prediction model is a random forest regression model.

[0033] Preferably, the clustered multi-source heterogeneous data of the area to be reconstructed is input into the trained settlement prediction model to obtain a predicted deformation map of the area to be reconstructed, specifically including:

[0034] The current clustered multi-source heterogeneous data of the region to be reconstructed is input into the trained random forest regression model to obtain the predicted deformation values ​​of multiple regression decision trees;

[0035] The mean of the predicted deformation values ​​of the multiple regression decision trees is calculated to obtain the predicted deformation map of the region to be reconstructed.

[0036] A second aspect of the present invention provides a system for reconstructing a continuous surface deformation map, comprising:

[0037] A region determination module is used to determine the region to be reconstructed based on the surface deformation map of the study area.

[0038] The model training module is used to train a land subsidence prediction model based on historical surface deformation data and historical clustered multi-source heterogeneous data of the study area.

[0039] The prediction module is used to input the current clustered multi-source heterogeneous data of the area to be reconstructed into the trained settlement prediction model to obtain the predicted deformation map of the area to be reconstructed.

[0040] A reconstruction module is used to fuse the surface deformation map and the predicted deformation map to obtain a continuous surface deformation map of the study area.

[0041] The method and system for reconstructing continuous surface deformation maps of the present invention have the following advantages compared with the prior art:

[0042] This invention proposes a multi-factor heterogeneous random forest regression model method to reconstruct (i.e., predict in the spatial domain) large-scale continuous surface deformation information. This method can fully utilize heterogeneous environmental factor data from various sources, reducing reliance on existing historical data. Simultaneously, the use of machine learning modeling increases the interpretability of the prediction results, making them more reliable and accurate. Furthermore, the use of K-Means clustering to cluster multi-source heterogeneous data can significantly reduce the impact of spatial heterogeneity of different source factors on the construction of the surface subsidence prediction model, improving modeling accuracy. This method is suitable for large-scale spatial deformation prediction, and the continuous surface deformation results obtained are beneficial for describing spatiotemporal surface deformation characteristics, explaining mechanisms, and making preventative decisions. Attached Figure Description

[0043] Figure 1 This is a simplified flowchart of the method for reconstructing a continuous surface deformation map according to an embodiment of the present invention;

[0044] Figure 2 The following is a deformation rate map of Xi'an City from 2012 to 2015 obtained using SBAS-InSAR technology in a specific embodiment of the present invention, wherein (a) is the deformation rate map of 2012, (b) is the deformation rate map of 2013, (c) is the deformation rate map of 2014, and (d) is the deformation rate map of 2015.

[0045] Figure 3 This is a flowchart of the reconstruction method corresponding to a specific embodiment of the present invention;

[0046] Figure 4This is a schematic diagram of the mesh discretization process in a specific embodiment of the present invention, wherein (a) is a schematic diagram before mesh discretization and (b) is a schematic diagram after mesh discretization;

[0047] Figure 5 In a specific embodiment of the present invention, the predicted deformation rate map of Xi'an City from 2012 to 2015 was obtained using the method of the present invention, wherein (a) is the predicted deformation rate map for 2012, (b) is the predicted deformation rate map for 2013, (c) is the predicted deformation rate map for 2014, and (d) is the predicted deformation rate map for 2015.

[0048] Figure 6 In the diagram, (a) is the deformation rate map obtained by InSAR in a specific embodiment of the present invention, (b) is the deformation rate map predicted for 2012-2015 in the area to be reconstructed using the method of the present invention, (c) is the continuous deformation map of the land surface after fusing the measured deformation rate map and the predicted deformation map, (d) to (f) are the deformation rate map, the predicted deformation rate map and the fused continuous deformation map of the land surface obtained by InSAR in region I in (c), respectively, (h) to (j) are the deformation rate map, the predicted deformation rate map and the fused continuous deformation map of the land surface obtained by InSAR in region II in (c), respectively, and (k) to (m) are the deformation rate map, the predicted deformation rate map and the fused continuous deformation map of the land surface obtained by InSAR in region III in (c), respectively.

[0049] Figure 7 The figures shown are comparisons of the prediction results of the random forest regression model before and after using K-Means clustering in a specific embodiment of the present invention. (a) to (c) are the predicted deformation rate map without K-Means clustering, the fused continuous surface deformation map, and the prediction accuracy map of the random forest regression model, respectively; (d) to (f) are the predicted deformation rate map with K-Means clustering, the fused continuous surface deformation map, and the prediction accuracy map of the random forest regression model, respectively.

[0050] Figure 8 The figures show a comparison between the reconstruction results of specific embodiments of the present invention and conventional methods. In the figures, (a), (d), and (h) are the deformation results reconstructed using the IDW method after removing existing settlement points in regions I to III, respectively; (b), (e), and (i) are the deformation results reconstructed using the ordinary Kriging method after removing existing settlement points in regions I to III, respectively; and (c), (f), and (j) are the deformation results reconstructed using the method of the present invention after removing existing settlement points in regions I to III, respectively.

[0051] Figure 9The figures show a comparison of the reconstruction accuracy of the selected method and Kriging and IDW in regions I-III in specific embodiments of the present invention, where (a) to (c) are comparisons of the reconstruction accuracy of the method of the present invention and Kriging and IDW in regions I-III, respectively; and (d) to (f) are prediction accuracy figures using the IDW method, using Kriging and using the method of the present invention, respectively. Detailed Implementation

[0052] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0053] A first aspect of this invention provides a method for reconstructing a continuous surface deformation map, such as... Figure 1 As shown, it includes:

[0054] Step 1: Based on the surface deformation map of the study area, determine the area to be reconstructed, specifically including:

[0055] Step 1.1: Divide the surface deformation map of the study area into multiple regions.

[0056] In this embodiment of the invention, a SAR image dataset of the study area is first acquired, and then the surface subsidence rate value is obtained by using differential interferometry short baseline set time series analysis technology (Small Baseline Subset InSAR, abbreviated as SBAS-InSAR) to obtain a surface deformation map.

[0057] In this embodiment of the invention, the surface deformation map of the study area is divided into multiple regions. Specifically, the surface deformation map can be divided into multiple continuous grids using meshes. The size of the grid depends primarily on the resolution of the subsequent impact factor dataset. In this embodiment, the average value of all surface subsidence rate values ​​within the grid cell is used as the grid measurement value.

[0058] For example, the surface deformation map is divided into multiple continuous grids of 100m*100m using Fishnet.

[0059] Step 1.2: Identify areas that do not contain surface deformation data as areas to be reconstructed.

[0060] For example, areas that do not contain surface deformation data are grids that do not contain surface deformation data, and grids that do not have surface subsidence rate values ​​are identified as areas to be reconstructed.

[0061] Following step 1, the following is also included:

[0062] Step S1: Obtain historical multi-source heterogeneous datasets. These datasets include historical data corresponding to the multi-source heterogeneous influence factors of the study area.

[0063] For example, the multi-source heterogeneous influencing factors include groundwater, rainfall, ground fissures, stratigraphic lithology, geomorphology, hydrogeology, engineering geology, land use type, soil type, GDP, DEM, etc.

[0064] To ensure the accuracy of subsequent model training, this embodiment of the invention projects the historical data of the aforementioned multi-source heterogeneous influencing factors onto the WGS84 coordinate system. Then, the historical data of vector factors such as groundwater and ground fissures are converted into raster data, and all raster datasets are resampled to a resolution of 100*100m. Furthermore, the categorical factors are subjected to hierarchical quantization processing, and then normalized simultaneously with the raster data.

[0065] Step S2: From the historical multi-source heterogeneous dataset, identify multiple historical heterogeneous data from different sources that have a correlation with surface deformation data greater than a preset first threshold, specifically including:

[0066] Step S21: Obtain the frequency ratio of each historical heterogeneous data in the surface deformation map from the historical multi-source heterogeneous dataset, specifically including:

[0067] Obtain the first ratio between the number of settlement areas affected by each type of historical heterogeneous data in the historical multi-source heterogeneous dataset and the total number of settlement areas;

[0068] Obtain the second ratio between the area occupied by each historical heterogeneous data in the surface deformation map and the total number of areas in the historical multi-source heterogeneous dataset;

[0069] Based on the first ratio and the second ratio, determine the frequency ratio of each type of historical heterogeneous data in the surface deformation map.

[0070] For example, when using a grid to divide the surface deformation map of the study area, the area in step S21 is the grid, for example: the number of subsidence areas is the number of subsidence grids.

[0071] In this embodiment of the invention, areas with a settlement rate of less than 5 mm / year are ignored. Multi-source heterogeneous factors and settlement areas are superimposed to calculate the number of settlement locations and the pixels associated with each category.

[0072] For example, the frequency ratio of each type of historical heterogeneous data in the surface deformation map is determined using the following formula (1):

[0073]

[0074] In the formula, FR represents the frequency ratio of each type of historical heterogeneous data in the surface deformation map of the study area, A is the number of surface subsidence grids affected by each type of historical heterogeneous data, A' is the total number of surface subsidence grids, B is the number of grids occupied by each type of historical heterogeneous data in the surface deformation map, and B' is the total number of grids in the surface deformation map.

[0075] Step S22: Based on the correlation between the frequency ratio and the surface subsidence rate value, determine the historical heterogeneous data from multiple different sources that have a correlation with the surface subsidence rate value greater than a preset first threshold.

[0076] In this embodiment of the invention, the method for calculating the correlation value can use conventional methods in the prior art. For example, the grey relational analysis method can be used to calculate the correlation value between the frequency ratio and the surface subsidence rate, screen out factors with a high correlation with surface subsidence, and remove factors with a low correlation.

[0077] Step S3: Cluster the heterogeneous data from multiple different sources that exceed the preset first threshold to obtain clustered multi-source heterogeneous data.

[0078] This invention clusters heterogeneous data from multiple different sources that exceed a preset first threshold. The clustering method used can be K-Means clustering, hierarchical clustering, SOM clustering, or FCM clustering, etc. The preferred embodiment of this invention uses the K-Means clustering method.

[0079] The process of clustering heterogeneous data from different sources using the K-Means clustering method is as follows: samples are iteratively grouped into K classes, and the cluster centroids are updated one by one to minimize the sum of distances d between each sample and the center or mean of its class, until the best clustering effect is obtained, where d is determined according to formula (2):

[0080]

[0081] In the formula, x ij For the i-th influencing factor of category j, c j Let be the center of the j-th class, and k be the number of clusters.

[0082] Once the clustering has stabilized, the optimal classification is determined according to the elbow rule formula, dividing the study area into several homogeneous regions.

[0083] Step 2: Train the surface subsidence prediction model based on historical surface deformation data and historical clustered multi-source heterogeneous data of the study area.

[0084] The settlement prediction model in this embodiment of the invention is a random forest regression model.

[0085] In this embodiment of the invention, during the training process, the clustered historical multi-source heterogeneous data and the historical surface deformation data of the study area are divided into a 70% training set and a 30% test set. Iterative testing is then used to determine the optimal parameters and construct a surface subsidence prediction model. The historical surface deformation data of the study area represents the historical surface subsidence rate values.

[0086] Step 3: Input the current clustered multi-source heterogeneous data of the area to be reconstructed into the trained settlement prediction model to obtain the predicted deformation map of the area to be reconstructed, specifically including:

[0087] Step 31: Input the current clustered multi-source heterogeneous data of the region to be reconstructed into the trained random forest regression model to obtain the predicted deformation map of multiple regression decision trees;

[0088] Step 32: Based on the idea of ​​integrated learning, calculate the mean of the predicted deformation values ​​of multiple regression decision trees to obtain the predicted deformation map of the area to be reconstructed, as shown in formula (3):

[0089]

[0090] In the formula: The predicted deformation map of the region to be reconstructed; h(x, θ) t (based on x and θ) t The output, x is the independent variable, θ t is an independent and identically distributed random vector; T is the number of regression decision trees.

[0091] The embodiments of the present invention utilize R 2 The accuracy of the model is evaluated using mean absolute error (MAE) and root mean square error (RMSE). 2 The conventional formulas for mean absolute error (MAE) and root mean square error (RMSE) are Equations (5) to (7), respectively.

[0092]

[0093]

[0094]

[0095] In the formula, y i This represents the true value of the deformation. The mean of the deformation values. This is the predicted value for deformation.

[0096] Step 4: Fuse the surface deformation map and the predicted deformation map to obtain a high-precision continuous surface deformation map of a large study area.

[0097] Furthermore, to avoid collinearity factor data affecting modeling accuracy, this embodiment of the invention includes the following step before step S3:

[0098] The correlation between any two historical heterogeneous data sets from multiple different sources is obtained when the correlation with the surface subsidence rate value is greater than a preset first threshold.

[0099] When the correlation is greater than the preset second threshold, one of the two corresponding historical heterogeneous data is selected to obtain multiple historical heterogeneous data from different sources after correlation processing.

[0100] Accordingly, step S3 is as follows:

[0101] Clustering is performed on historical heterogeneous data from multiple different sources after correlation processing to obtain clustered multi-source heterogeneous data.

[0102] This invention calculates the correlation between any two historical heterogeneous data sets using either the Pearson correlation coefficient method or the Spearman rank correlation coefficient method. The preferred embodiment of this invention uses the Spearman rank correlation coefficient method. The Spearman correlation coefficient method is used to describe the correlation between two vectors, where the two variables are two historical heterogeneous data sets. The method for calculating the Spearman correlation coefficient of two n-dimensional vectors x and y is as follows:

[0103]

[0104] In the formula R i and S i The order of the observed value i after sorting the representative vectors x and y. and This represents the average order of vectors x and y, where N is the total number of i, and d i =R i -S i ρ represents the difference in the order of i among two variables. In the Spearman correlation coefficient, ρ... s It is a real number in the range [-1, 1]. When the correlation coefficient is greater than zero, the two variables are positively correlated; otherwise, they are negatively correlated.

[0105] In this embodiment of the invention, the second threshold is set to 0.8.

[0106] This invention proposes a method for reconstructing a continuous deformation map of the land surface using a multi-factor random forest regression model. This method comprehensively considers multiple influencing factors (multi-source heterogeneous data) affecting land surface subsidence and the spatial heterogeneity of each factor, and reconstructs a large-scale continuous deformation field of the land surface with high accuracy.

[0107] A second aspect of the present invention provides a reconstruction system for a continuous deformation map of the earth's surface, including a region determination module, a model training module, a prediction module, and a reconstruction module.

[0108] The region determination module is used to determine the region to be reconstructed based on the surface deformation map of the study area.

[0109] The model training module is used to train a land subsidence prediction model based on historical surface deformation data and historical clustered multi-source heterogeneous data of the study area;

[0110] The prediction module is used to input the current clustered multi-source heterogeneous data of the area to be reconstructed into the trained settlement prediction model to obtain the predicted deformation map of the area to be reconstructed.

[0111] The reconstruction module is used to fuse the surface deformation map and the predicted deformation map to obtain a continuous surface deformation map of the study area.

[0112] This invention utilizes a random forest regression model method that considers multi-factor heterogeneity to reconstruct (i.e., predict in the spatial domain) large-scale continuous surface deformation information. It fully leverages heterogeneous environmental factor data from different sources, reducing reliance on existing historical data. Simultaneously, the use of machine learning modeling increases the interpretability of the prediction results, making them more reliable and accurate. Furthermore, the use of K-Means clustering to cluster multi-source heterogeneous data significantly reduces the impact of spatial heterogeneity of different source factors on the construction of the surface subsidence prediction model, improving modeling accuracy. This method is suitable for large-scale spatial deformation prediction, and the continuous surface deformation results obtained are beneficial for describing spatiotemporal surface deformation characteristics, explaining mechanisms, and making preventative decisions.

[0113] The method of the present invention will now be described in detail with more specific embodiments.

[0114] This embodiment selects Xi'an City, located in the northern Qinling Mountains of my country. Xi'an has a complex topography, with as many as 10 loess depressions formed by tectonic activity and several active fissures, among which the Weihe Fault and the Lintong-Chang'an Fault are the most typical. Due to rapid population growth and a significant increase in industrial water consumption, surface water cannot meet the demand, making groundwater an important source of water supply for Xi'an. Long-term over-exploitation of groundwater is one of the causes of land subsidence. With the rapid economic development and urban expansion of Xi'an, the frequent construction of underground projects and high-rise buildings has become another cause of land subsidence. The combined effects of natural factors such as geological fissures and fault zones, along with anthropogenic factors such as groundwater extraction, have led to regional land subsidence in parts of Xi'an.

[0115] In this embodiment, to obtain the deformation characteristics of the study area, we used 40 scenes of German TerraSAR-X data from January 2012 to May 2015 as the baseline dataset and obtained the surface subsidence rate values ​​of the study area using SBAS-InSAR technology. We also obtained five dynamic influencing factors: confined aquifer level, confined aquifer change, groundwater level, groundwater change, and rainfall; and nine static influencing factors: ground fissures, stratigraphic lithology, geomorphology, hydrogeology, engineering geology, land use type, soil type, GDP, and DEM.

[0116] The annual average deformation rate maps for Xi'an City in 2012, 2013, 2014, and 2015 were calculated using SBAS-InSAR technology, as shown below. Figure 2 As shown. However, many areas of missing measurement information exist in the measurement results, making it difficult to analyze spatiotemporal characteristics. Therefore, a method for reconstructing continuous surface deformation maps of Xi'an City using a multi-factor random forest regression model (also known as the K-RFR method) is employed. The process is as follows: Figure 3 The average deformation rate map was discretized into 100m*100m grid cells. The measured area and the area without measurement information contained 47810 and 42496 cells respectively, meaning that approximately 47% of the area needed to be reconstructed to obtain a continuous deformation map. A schematic diagram of the grid discretization is shown below. Figure 4 .

[0117] Figure 5 This is a continuous surface deformation map reconstructed using the method selected in this invention. The reconstructed area achieves a spatial resolution of 100m, and the predicted results show good spatial continuity with the measured results. To observe the details of the reconstructed deformation results, Figure 6 Three typical areas were shown. Figure 6 (a) shows the deformation rate obtained by InSAR technology. Figure 6 (b) shows the deformation rates predicted for 2012–2015 in a region lacking measurement information using the method selected in this invention. Figure 6 (c) shows the result after fusing the measurement results and the prediction results. Figure 6 (dm) shows the deformation details of three typical regions.

[0118] To further demonstrate that K-Means clustering analysis improves the accuracy of foundation deformation prediction models, the absolute error and absolute error distribution between the prediction results of the random forest regression model before and after clustering and the InSAR measurements were calculated, as follows: Figure 7As shown, both clustering methods can reconstruct surface deformation well. The absolute error of most prediction results is less than 6 mm, and the error is less than 10% of the settlement rate, which meets the prediction requirements. Compared with the prediction results of the model constructed before clustering, the standard deviation of the prediction results of the method selected in this invention decreased from 2.352 mm to 1.782 mm, verifying that the method selected in this invention has higher prediction accuracy.

[0119] Figure 8 To compare the deformation results reconstructed by the method selected in this invention with the cumulative deformation results reconstructed by IDW and ordinary Kriging after removing existing settlement points in regions I-III, the first column represents the IDW method, the second column represents the ordinary Kriging method, and the third column represents the method selected in this invention. Compared with traditional interpolation methods such as IDW and ordinary Kriging, the method selected in this invention can better reflect the spatial variation of settlement. Figure 9 This is a comparison of the accuracy of the reconstruction results of the method selected in this invention with those of Kriging and IDW in regions I-III. High-coherence points removed are used as validation data, compared with IDW(R... 2 =0.63) and ordinary Kriging(R 2 Compared to (=0.69), the method selected in this invention predicts (R... 2 =0.94) The results showed better agreement with InSAR monitoring results than the two spatial interpolation methods.

[0120] The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A method for reconstructing a continuous surface deformation map, characterized in that, include: Step 1: Determine the area to be reconstructed based on the surface deformation map of the study area; Step 2: Train the surface subsidence prediction model based on historical surface deformation data and historical clustered multi-source heterogeneous data of the study area; Step 3: Input the current clustered multi-source heterogeneous data of the area to be reconstructed into the trained settlement prediction model to obtain the predicted deformation map of the area to be reconstructed. Step 4: Fuse the surface deformation map and the predicted deformation map to obtain a continuous surface deformation map of the study area; The steps preceding step 2 also include: Obtain historical multi-source heterogeneous datasets; From the historical multi-source heterogeneous dataset, determining multiple historical heterogeneous data from different sources whose correlation with surface deformation data is greater than a preset first threshold specifically includes: obtaining the frequency ratio of each type of historical heterogeneous data in the historical multi-source heterogeneous dataset in the surface deformation map; determining multiple historical heterogeneous data from different sources whose correlation with surface subsidence rate values ​​is greater than a preset first threshold based on the correlation between the frequency ratio and the surface subsidence rate value; obtaining the frequency ratio of each type of historical heterogeneous data in the historical multi-source heterogeneous dataset in the surface deformation map specifically includes: obtaining a first ratio between the number of subsidence areas affected by each type of historical heterogeneous data in the historical multi-source heterogeneous dataset and the total number of subsidence areas; obtaining a second ratio between the number of areas occupied by each type of historical heterogeneous data in the historical multi-source heterogeneous dataset in the surface deformation map and the total number of areas; determining the frequency ratio of each type of historical heterogeneous data in the surface deformation map based on the first ratio and the second ratio; Clustering is performed on historical heterogeneous data from multiple different sources that exceed a preset first threshold to obtain clustered multi-source heterogeneous data.

2. The method for reconstructing a continuous surface deformation map according to claim 1, characterized in that, Step 1 specifically includes: The surface deformation map of the study area is divided into multiple regions; Areas that do not contain surface deformation data are identified as areas to be reconstructed.

3. The method for reconstructing a continuous surface deformation map according to claim 1, characterized in that, Clustering is performed on heterogeneous data from multiple different sources that exceed a preset first threshold, specifically including: The K-Means clustering method is used to cluster heterogeneous data from multiple different sources that exceed a preset first threshold.

4. The method for reconstructing a continuous surface deformation map according to claim 1, characterized in that, Before clustering historical heterogeneous data from multiple different sources that exceed a preset first threshold, the following steps are also included: The correlation between any two historical heterogeneous data sets from multiple different sources is obtained when the correlation with the surface subsidence rate value is greater than a preset first threshold. When the correlation is greater than the preset second threshold, one of the two corresponding historical heterogeneous data is selected to obtain multiple historical heterogeneous data from different sources after correlation processing. Accordingly, clustering is performed on multiple historical heterogeneous data from different sources that exceed a preset first threshold, specifically as follows: Clustering is performed on historical heterogeneous data from multiple different sources after correlation processing.

5. The method for reconstructing a continuous surface deformation map according to claim 1, characterized in that, The settlement prediction model is a random forest regression model.

6. The method for reconstructing a continuous surface deformation map according to claim 5, characterized in that, The clustered multi-source heterogeneous data of the area to be reconstructed is input into the trained settlement prediction model to obtain the predicted deformation map of the area to be reconstructed, specifically including: The current clustered multi-source heterogeneous data of the region to be reconstructed is input into the trained random forest regression model to obtain the predicted deformation values ​​of multiple regression decision trees; The mean of the predicted deformation values ​​of the multiple regression decision trees is calculated to obtain the predicted deformation map of the region to be reconstructed.

7. A system for reconstructing a continuous surface deformation map based on the reconstruction method of any one of claims 1-6, characterized in that, include: A region determination module is used to determine the region to be reconstructed based on the surface deformation map of the study area. The model training module is used to train a land subsidence prediction model based on historical surface deformation data and historical clustered multi-source heterogeneous data of the study area. The prediction module is used to input the current clustered multi-source heterogeneous data of the area to be reconstructed into the trained settlement prediction model to obtain the predicted deformation map of the area to be reconstructed. A reconstruction module is used to fuse the surface deformation map and the predicted deformation map to obtain a continuous surface deformation map of the study area.

Images