A ground surface deformation monitoring and predicting method

Abstract

The present application provides a kind of ground surface deformation monitoring and prediction method, comprising: obtaining the InSAR time series ground surface deformation data set of research area, and processing it, obtain the environmental factor data set of research area, and processing it, the environmental factor data set after processing and the InSAR time series ground surface deformation data set after processing are merged, obtain model data set, build Autoformer model, Autoformer model is trained by model data set, adjust model hyperparameter, obtain best model, select the study point set of the largest ground surface deformation change in research area, input best model and carry out prediction warning.The ground surface deformation monitoring and prediction method provided by the present application does not need ground and manual cooperation, can give full play to the advantages of satellite monitoring, can carry out ground surface deformation monitoring to the full range of research area, and can realize prediction, can reduce accident incidence and property loss rate.

Description

Technical Field

[0001] This invention relates to the field of surface deformation monitoring technology, and in particular to a method for monitoring and predicting surface deformation. Background Technology

[0002] In the monitoring of surface deformation, traditional geodetic techniques, such as GPS and manual leveling instruments, are easily limited by low spatial resolution and high cost. In addition, in the inspection of ground features, traditional techniques often require a large number of professional personnel and heavy equipment, resulting in long monitoring time and high monitoring costs.

[0003] Existing technologies also utilize Multi-Temporal Interferometric Synthetic Aperture Radar (MT-InSAR) to achieve large-scale, high-precision, and low-cost monitoring. One typical MT-InSAR technique is Permanent Scatter Radar Interference Synthetic Aperture Radar (PS-InSAR). This method selects time-stable points such as buildings and exposed rocks, which exhibit high stability and reflectivity across multiple SAR images. However, current PS-InSAR applications often fail to integrate it with long-term time-series prediction and analysis frameworks, thus failing to predict surface subsidence in densely packed areas and neglecting to consider environmental factors influencing surface deformation. Therefore, designing a surface deformation monitoring and prediction method based on Autoformer and PS-InSAR is essential. Summary of the Invention

[0004] The purpose of this invention is to provide a method for monitoring and predicting land surface deformation that does not require ground or human intervention, fully leverages the advantages of satellite monitoring, can monitor land surface deformation across the entire study area, and can make predictions, thereby reducing the accident rate and property loss rate.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for monitoring and predicting land surface deformation includes the following steps:

[0007] Step 1: Obtain the InSAR time-series surface deformation dataset for the study area;

[0008] Step 2: Process the acquired InSAR time series surface deformation dataset;

[0009] Step 3: Obtain the environmental factor dataset for the study area;

[0010] Step 4: Process the environmental factor dataset, and merge the processed environmental factor dataset with the processed InSAR time series surface deformation dataset to obtain the model dataset;

[0011] Step 5: Build the Autoformer model, train the Autoformer model using the model dataset, adjust the model hyperparameters, and obtain the optimal model;

[0012] Step 6: Select the set of study points with the greatest surface deformation changes within the study area, and input them into the best model for prediction and early warning.

[0013] Optionally, in step 1, the InSAR time-series surface deformation dataset of the study area is obtained, specifically as follows:

[0014] Acquire n Sentinel-1A SAR images covering the study area. Based on the spatiotemporal baseline parameters and Doppler center frequency, select one SAR image as the master image and the rest as auxiliary images. Perform precise registration and interferometric processing on the master and auxiliary images to generate N-1 differential interferograms. Acquire external DEM data and satellite orbit data. Register the external DEM data with the image coverage area. Perform orbital offset vector correction to remove plane effects using satellite orbit data. Sample the DEM data into the coordinate system of the SAR image. Identify PS points using the amplitude deviation method. Set a reasonable amplitude deviation threshold, extract PS candidate points in the study area, and construct a pixel network. The amplitude deviation method confirms PS points by evaluating the signal-to-noise ratio of image pixels. The phase signal-to-noise ratio can be measured by the amplitude deviation exponent, i.e.:

[0015]

[0016] In the formula, σ A and μ A These represent the standard deviation and mean amplitude of the SAR image, respectively, and D is set as follows: A The threshold is x. When the amplitude deviation index at a given location is less than x, it is determined to be a PS point; otherwise, it is determined not to be a PS point. A deformation inversion model is established as follows:

[0017] ψ x,i =W{φ D,x,i +φ A,x,i +Δφ S,x,i +Δφ θ,x,i +Δφ N,x,i}

[0018] In the formula, W is the phase winding operator, φ D,x,i The phase caused by the ground phase, φ A,x,i For atmospheric delay error phase, Δφ S,x,i For the satellite orbital error phase, Δφ θ,x,i For the residual terrain phase, Δφ N,x,iTo address the registration error, specifically the noise phase caused by thermal noise decorrelation, a deformation inversion model was established. Phase parameters of adjacent PS points were estimated, residual elevation, linear deformation, and atmospheric phase parameters were calculated. The fit between the model's calculated parameters and the interferometric phase of the pixels was tested to demonstrate the accuracy of the model's calculations. After atmospheric phase removal, PS candidate points were reselected, and deformation analysis was performed again. Finally, the residuals were decomposed to obtain nonlinear deformation information, resulting in a time-series surface deformation dataset. After geocoding, an InSAR time-series surface deformation dataset was obtained.

[0019] Optionally, in step 2, the acquired InSAR time-series surface deformation dataset is processed, specifically as follows:

[0020] The acquired InSAR time series surface deformation dataset is cropped and smoothed according to the study area to obtain the length of the time series dataset and convert the file format. The key points of the study area, namely the set of points with the greatest deformation, are identified to obtain the processed InSAR time series surface deformation dataset.

[0021] Optionally, in step 3, the environmental factor dataset for the study area is obtained, specifically as follows:

[0022] An environmental factors dataset is constructed by acquiring atmospheric driving dataset, surface temperature analysis dataset, soil moisture dataset, soil temperature analysis dataset, and soil relative humidity dataset in NetCDF format.

[0023] Optionally, in step 4, the environmental factor dataset is processed, and the processed environmental factor dataset and the processed InSAR time series surface deformation dataset are merged to obtain the model dataset, specifically:

[0024] The environmental factor dataset in NetCDF format was converted to CSV format using Pandas and NumPy libraries based on Python. Environmental factor parameters were extracted from the dataset, and the dataset was divided according to a latitude and longitude grid of the study area. The extracted environmental factor parameters were then standardized. After standardization, the dataset underwent a data format review to remove data noise that did not conform to the standard format. Outlier correction was then performed using the average value of each environmental factor parameter to fill in missing values, resulting in a processed environmental factor dataset. The distance between each study point and the meteorological station collecting the environmental factor parameters within the study area was calculated. For each study point, the environmental factor parameters collected by the nearest meteorological station were selected as a reference. Clustering was performed on the study points, and the environmental factor dataset and the InSAR surface deformation time series dataset were merged to obtain the model dataset.

[0025] According to specific embodiments provided by the present invention, the following technical effects are disclosed: The surface deformation monitoring and prediction method provided by the present invention includes acquiring an InSAR time-series surface deformation dataset of a study area, processing the acquired InSAR time-series surface deformation dataset, acquiring an environmental factor dataset of the study area, processing the environmental factor dataset, merging the processed environmental factor dataset and the processed InSAR time-series surface deformation dataset to obtain a model dataset, building an Autoformer model, training the Autoformer model using the model dataset, adjusting the model hyperparameters, obtaining the optimal model, selecting the set of study points with the largest surface deformation changes in the study area, and inputting the optimal model for prediction and early warning; This method combines Autoformer and PS-InSAR and fully considers the environmental factors affecting surface deformation, enabling large-scale and high-precision monitoring of surface deformation, effectively reducing the economic investment in surface deformation monitoring, providing a new approach for preventing surface subsidence disasters, using PS-InSAR to monitor surface deformation can achieve millimeter-level wide-area monitoring with high accuracy, and using Autoformer, its prediction model accuracy far exceeds that of similar prediction models. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1This is a schematic diagram of the surface deformation monitoring and prediction method according to an embodiment of the present invention;

[0028] Figure 2 This is a schematic diagram illustrating the process of acquiring InSAR time-series surface deformation datasets.

[0029] Figure 3 A comparison chart of the evaluation indicators for each model;

[0030] Figure 4 A comparison chart of evaluation metrics for the Autoformer experimental models at various research sites. Detailed Implementation

[0031] The technical solutions of the embodiments 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, and 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.

[0032] The purpose of this invention is to provide a method for monitoring and predicting land surface deformation that does not require ground or human intervention, fully leverages the advantages of satellite monitoring, can monitor land surface deformation across the entire study area, and can make predictions, thereby reducing the accident rate and property loss rate.

[0033] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] like Figure 1 As shown, the surface deformation monitoring and prediction method provided in this embodiment of the invention includes the following steps:

[0035] Step 1: Obtain the InSAR time-series surface deformation dataset for the study area;

[0036] Step 2: Process the acquired InSAR time series surface deformation dataset;

[0037] Step 3: Obtain the environmental factor dataset for the study area;

[0038] Step 4: Process the environmental factor dataset, and merge the processed environmental factor dataset with the processed InSAR time series surface deformation dataset to obtain the model dataset;

[0039] Step 5: Build the Autoformer model, train the Autoformer model using the model dataset, adjust the model hyperparameters, and obtain the optimal model;

[0040] Step 6: Select the set of study points with the greatest surface deformation changes within the study area, and input them into the best model for prediction and early warning;

[0041] Surface deformation can be monitored using model datasets, and surface deformation can be predicted using the best model.

[0042] In step 1, the InSAR time-series surface deformation dataset of the study area is obtained, specifically as follows:

[0043] n Sentinel-1A SAR images covering the study area were acquired. Based on the spatiotemporal baseline parameters and Doppler center frequency, one SAR image was selected as the master image, and the rest as auxiliary images. The master and auxiliary images were precisely registered and interferometrically processed to generate N-1 differential interferograms. The phase difference between adjacent PS points was obtained through the differential interferograms. A phase difference model solution for adjacent PS points was established, and the model parameters were calculated to obtain surface subsidence information. External DEM data and satellite orbit data were acquired. The external DEM data was registered with the image coverage area. Orbital offset vector correction was performed using satellite orbit data to remove plane effects. The DEM data was sampled into the coordinate system of the SAR image. PS points were identified using the amplitude deviation method. A reasonable amplitude deviation threshold was set, and candidate PS points for the study area were extracted, constructing a pixel network. The amplitude deviation method confirms PS points by evaluating the signal-to-noise ratio of image pixels. The phase signal-to-noise ratio can be measured by the amplitude deviation exponent, i.e.:

[0044]

[0045] In the formula, σ A and μ A These represent the standard deviation and mean amplitude of the SAR image, respectively, and D is set as follows: A The threshold is x. When the amplitude deviation index at a given location is less than x, it is determined to be a PS point; otherwise, it is determined not to be a PS point. A deformation inversion model is established using the phase difference caused by the ground phase, atmospheric delay error, satellite orbital error phase, residual terrain phase, registration error, and phase unwrapping operator. The deformation inversion model is used to estimate the phase parameters of adjacent PS points and the model parameters. The deformation inversion model is as follows:

[0046] ψ x,i =W{φ D,x,i +φ A,x,i +Δφ S,x,i +Δφ θ,x,i +Δφ N,x,i}

[0047] In the formula, W is the phase winding operator, φ D,x,i The phase caused by the ground phase, φ A,x,i For atmospheric delay error phase, ΔφS,x,i For the satellite orbital error phase, Δφ θ,x,i For the residual terrain phase, Δφ N,x,i To address registration errors, specifically the noise phase caused by thermal noise decorrelation, a deformation inversion model was established. Phase parameters of adjacent PS points were estimated, residual elevation, linear deformation, and atmospheric phase parameters were calculated. The fit between the model's calculated parameters and the pixel interference phase was tested to verify the accuracy of the model's solutions. After atmospheric phase removal, new PS candidate points were selected, and deformation analysis was performed again. Finally, the residuals were decomposed to obtain nonlinear deformation information, resulting in a time-series surface deformation dataset. After geocoding, an InSAR time-series surface deformation dataset was obtained. The specific flowchart is shown below. Figure 2 As shown.

[0048] In step 2, the acquired InSAR time-series surface deformation dataset is processed, specifically as follows:

[0049] The acquired InSAR time series surface deformation dataset is cropped and smoothed according to the study area to obtain the length of the time series dataset and convert the file format. The key points of the study area, namely the set of points with the greatest deformation, are identified to obtain the processed InSAR time series surface deformation dataset.

[0050] In step 3, the environmental factor dataset for the study area is obtained, specifically as follows:

[0051] The following data products in NetCDF format, provided by the China Meteorological Administration, were used to construct an environmental factors dataset: Atmospheric Driving Dataset (ADDS), Surface Temperature Analysis Dataset (STADS), Soil Moisture Dataset (SMDS), Soil Temperature Analysis Dataset (STADS), and Soil Relative Humidity Dataset (SRHDS). By comparing the root mean square error, bias, and correlation coefficient of each parameter in this dataset, it was confirmed that the environmental factors dataset has high spatiotemporal resolution and is of superior quality to similar international datasets. Specific data quality parameters are shown in Table 1.

[0052] Table 1. Description of Environmental Factor Data Quality

[0053]

[0054]

[0055] In step 4, the environmental factor dataset is processed, and the processed environmental factor dataset and the processed InSAR time series surface deformation dataset are merged to obtain the model dataset, specifically:

[0056] To construct an accurate and effective environmental factor dataset, this invention preprocesses the original environmental factor dataset, specifically including data format conversion, data rationality processing, data denoising, and outlier supplementation.

[0057] The data format conversion involved converting the NetCDF format environmental factor dataset into a CSV format environmental factor dataset using the Pandas and Numpy libraries based on the Python language, and then extracting the environmental factor parameters from the dataset.

[0058] The data rationality processing is as follows: the environmental factor dataset is divided according to the latitude and longitude grid of the study area, and the extracted environmental factor parameters are standardized in data format. The environmental factor data used in this invention is 1-day resolution monitoring data. In order to resolve the contradiction between the slow change of surface deformation and the excessively fast monitoring frequency of environmental factors, the average value processing of each environmental factor data in this invention is performed to obtain the daily average environmental factor parameters, remove data redundancy, and improve data rationality.

[0059] Data denoising involves: after standardization, reviewing the data format of the environmental factor dataset and removing data noise that does not conform to the standard format;

[0060] Outlier padding is performed as follows: After removing data noise, outliers are padded in the environmental factor dataset by using the average value of each environmental factor parameter to fill in the missing values ​​in the environmental factor dataset, resulting in a processed environmental factor dataset.

[0061] The distance between each study point within the study area and the environmental factor meteorological station that collected the environmental factor parameters was calculated. For each study point, the environmental factor parameters collected by the nearest environmental factor meteorological station were selected as a reference. The study points were then clustered and the environmental factor dataset and the InSAR surface deformation time series dataset were merged to obtain the model dataset.

[0062] This invention also compares and analyzes the accuracy of Autoformer, Transformer (an encoder-decoder model based on an attention mechanism), Informer (a long-term sequence prediction model), and Reformer (an efficient encoder-decoder model), ultimately confirming the effectiveness and accuracy of Autoformer in this research problem.

[0063] This invention evaluates the accuracy of the model using five accuracy metrics: root square error (MSE), mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), and mean square percentage error (MSPE). The specific formulas are as follows:

[0064]

[0065]

[0066]

[0067]

[0068]

[0069] Among them, y t This represents the monitored value of surface deformation at time t. The value of the predicted surface deformation at time t is represented by N, where N is the size of the test set. A comparison of the evaluation metrics for each model is shown in the figure below. Figure 3 As shown in the figure, the comparison of evaluation indicators of the Autoformer experimental model at each research point is as follows: Figure 4 As shown.

[0070] Autoformer is a deep learning network model based on a deep decomposition architecture and autocorrelation mechanism. It addresses the challenges of handling complex time patterns and achieving high computational efficiency in long-term series forecasting by using progressive decomposition and sequence-level connections to significantly improve the efficiency of long-term series forecasting. The Autoformer network structure includes an internal concatenated decomposition block, an autocorrelation mechanism, and corresponding decoders and encoders. The encoder eliminates the long-term trend cycle part through the sequence decomposition block and focuses on seasonal pattern modeling. The decoder gradually accumulates the trend part extracted from the hidden variables. The encoder-decoder pair utilizes the past seasonal information in the encoder for autocorrelation.

[0071] This invention uses PS-InSAR technology to monitor surface deformation, enabling millimeter-level monitoring and prediction of surface deformation. It can accurately monitor surface deformation in the study area and reduce property loss rate.

[0072] This invention realizes a method and process from data processing to deformation monitoring, key point early warning and key point prediction. It can be applied to various areas with dense surface features, such as the monitoring, inspection and prediction of urban infrastructure. When using it, users can determine the prediction duration according to their own needs. It takes into account the impact of environmental factors on surface deformation, and the prediction results are more accurate and more convincing.

[0073] This invention provides a method for monitoring and predicting land surface deformation. The method includes acquiring an InSAR time-series land surface deformation dataset of the study area, processing the acquired InSAR time-series land surface deformation dataset, acquiring an environmental factor dataset of the study area, processing the environmental factor dataset, and merging the processed environmental factor dataset and the processed InSAR time-series land surface deformation dataset to obtain a model dataset. An Autoformer model is then built, trained using the model dataset, and the model's hyperparameters are adjusted to obtain the optimal model. Finally, a set of study points with the largest land surface deformation changes within the study area is selected and input into the optimal model for prediction and early warning. This method combines Autoformer and PS-InSAR and fully considers the environmental factors affecting land surface deformation, enabling large-scale, high-precision monitoring of land surface deformation. It effectively reduces the economic investment in land surface deformation monitoring, providing a new approach for preventing land subsidence disasters. Using PS-InSAR to monitor land surface deformation achieves millimeter-level wide-area monitoring with high accuracy. The Autoformer prediction model's accuracy far exceeds that of similar prediction models.

[0074] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for monitoring and predicting surface deformation, characterized in that, Includes the following steps: Step 1: Obtain the InSAR time-series surface deformation dataset for the study area; Step 2: Process the acquired InSAR time series surface deformation dataset; Step 3: Obtain the environmental factor dataset for the study area; Step 4: Process the environmental factor dataset, and merge the processed environmental factor dataset with the processed InSAR time series surface deformation dataset to obtain the model dataset; Step 5: Build the Autoformer model, train the Autoformer model using the model dataset, adjust the model hyperparameters, and obtain the optimal model; Step 6: Select the set of study points with the greatest surface deformation changes within the study area, and input them into the best model for prediction and early warning; In step 1, the InSAR time-series surface deformation dataset of the study area is obtained, specifically as follows: Acquire n Sentinel-1A SAR images covering the study area. Based on the spatiotemporal baseline parameters and Doppler center frequency, select one SAR image as the master image and the rest as auxiliary images. Perform precise registration and interferometric processing on the master and auxiliary images to generate N-1 differential interferograms. Acquire external DEM data and satellite orbit data. Register the external DEM data with the image coverage area. Perform orbital offset vector correction to remove plane effects using satellite orbit data. Sample the DEM data into the coordinate system of the SAR image. Identify PS points using the amplitude deviation method. Set a reasonable amplitude deviation threshold, extract PS candidate points in the study area, and construct a pixel network. The amplitude deviation method confirms PS points by evaluating the signal-to-noise ratio of image pixels. The phase signal-to-noise ratio can be measured by the amplitude deviation exponent, i.e.: In the formula, and The standard deviation and mean amplitude of the SAR image are set respectively. The threshold is x. When the amplitude deviation index at a given location is less than x, it is determined to be a PS point; otherwise, it is determined not to be a PS point. A deformation inversion model is established as follows: In the formula, W is the phase winding operator. The phase caused by the ground phase, For atmospheric delay error phase, For satellite orbital error phase, For residual topographic phase, To address the registration error, specifically the noise phase caused by thermal noise decorrelation, a deformation inversion model was established. Phase parameters of adjacent PS points were estimated, residual elevation, linear deformation, and atmospheric phase parameters were calculated. The fitting of the model's calculated parameters to the interferometric phase of the pixels was tested to demonstrate the accuracy of the model's calculated values. After atmospheric phase removal, PS candidate points were reselected, and deformation analysis was performed again. Finally, the residuals were decomposed to obtain nonlinear deformation information, and a time-series surface deformation dataset was acquired. After geocoding, an InSAR time-series surface deformation dataset was obtained. In step 3, the environmental factor dataset for the study area is obtained, specifically as follows: The atmospheric driving dataset, surface temperature analysis dataset, soil moisture dataset, soil temperature analysis dataset, and soil relative humidity dataset in NetCDF format are used to form the environmental factors dataset. In step 4, the environmental factor dataset is processed, and the processed environmental factor dataset and the processed InSAR time series surface deformation dataset are merged to obtain the model dataset, specifically: The environmental factor dataset in NetCDF format was converted to CSV format using Pandas and NumPy libraries based on Python. Environmental factor parameters were extracted from the dataset, and the dataset was divided according to a latitude and longitude grid of the study area. The extracted environmental factor parameters were then standardized. After standardization, the dataset underwent a data format review to remove data noise that did not conform to the standard format. Outlier correction was then performed using the average value of each environmental factor parameter to fill in missing values, resulting in a processed environmental factor dataset. The distance between each study point and the meteorological station collecting the environmental factor parameters within the study area was calculated. For each study point, the environmental factor parameters collected by the nearest meteorological station were selected as a reference. Clustering was performed on the study points, and the environmental factor dataset and the InSAR surface deformation time series dataset were merged to obtain the model dataset.

2. The method for monitoring and predicting surface deformation according to claim 1, characterized in that, In step 2, the acquired InSAR time-series surface deformation dataset is processed, specifically as follows: The acquired InSAR time series surface deformation dataset is cropped and smoothed according to the study area to obtain the length of the time series dataset and convert the file format. The key points of the study area, namely the set of points with the greatest deformation, are identified to obtain the processed InSAR time series surface deformation dataset.

Images