A Ground Settlement Prediction Method Based on Spatiotemporal Sequence Data

Abstract

The present application provides a ground settlement prediction method based on spatio-temporal sequence data, including: obtaining SAR images, orbital data, and DEM files within a preset time period in a preset geographical area; based on the SAR images, the orbital data, and the DEM files, obtaining n time-series settlement data sets arranged in chronological order, each time-series settlement data set including historical settlement values of m prediction points; predicting the settlement value of the future time node of prediction point i based on the historical settlement value of prediction point i and the historical settlement values of prediction points having spatial correlation with prediction point i, where the value of i ranges from 1 to m. The present invention can improve the rationality and accuracy of prediction.

Description

Technical Field

[0001] This application relates to the field of land subsidence, and specifically relates to a method for predicting land subsidence based on spatio-temporal sequence data. Background Art

[0002] Land subsidence refers to a geological phenomenon in which underground loose rock layers are consolidated and compressed due to natural factors or human engineering activities, resulting in a decrease in ground elevation within a certain area, and it belongs to a slow-changing geological disaster. Among them, as an important means of preventing and controlling disasters, the monitoring and prediction technologies of land subsidence have been greatly developed.

[0003] Traditional time series prediction models mainly use time series data to predict land subsidence. However, geographical things or attributes are interrelated in spatial distribution, and since time series data usually does not contain spatial information, based on traditional time series prediction models, accurate land subsidence prediction results cannot be obtained. Summary of the Invention

[0004] In view of the above technical problems, the technical solution adopted in this application is as follows:

[0005] An embodiment of the present invention provides a method for predicting land subsidence based on spatio-temporal sequence data, and the method includes the following steps:

[0006] S100, obtaining SAR images, orbit data, and DEM files within a preset time period of a preset geographical area;

[0007] S200, based on the SAR images, the orbit data, and the DEM files, obtaining n time series settlement data sets arranged in chronological order, and each time series settlement data set includes historical settlement values of m prediction points;

[0008] S300, predicting the settlement value of the future time node of prediction point i based on the historical settlement value of prediction point i and the historical settlement values of prediction points that have spatial correlation with prediction point i, where the value of i ranges from 1 to m.

[0009] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, in which at least one instruction or at least one program segment is stored, and the at least one instruction or the at least one program segment is loaded and executed by a processor to implement the foregoing method.

[0010] An embodiment of the present invention also provides an electronic device, which is characterized by including a processor and the foregoing non-transitory computer-readable storage medium.

[0011] An embodiment of the present invention also provides a computer program product, including a computer program, which is characterized in that the computer program is executed by a processor to implement the foregoing method.

[0012] The ground settlement prediction method based on spatio-temporal sequence data provided by the embodiments of the present application can more accurately and effectively predict ground settlement and obtain more accurate prediction results because it considers the influence of spatial correlation on surface settlement prediction, that is, it considers the spatio-temporal correlation of data and can capture both the temporal and spatial correlations of spatio-temporal data. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 It is a flowchart of a ground settlement prediction method based on spatio-temporal sequence data provided by the embodiments of the present application;

[0015] Figure 2 It is a schematic diagram of interferometric processing of SAR images;

[0016] Figure 3 It is a schematic diagram of time series deformation estimation of SAR images after interferometric processing;

[0017] Figure 4 It is a schematic diagram of the calculation of the gate structure;

[0018] Figures 5(a) to 5(c) It is a schematic diagram of the convolution kernel and the processing using the convolution kernel. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0020] Figure 1 It is a flowchart of a ground settlement prediction method based on spatio-temporal sequence data provided by the embodiments of the present application. As Figure 1 shown, the ground settlement prediction method based on spatio-temporal sequence data provided by the embodiments of the present invention includes the following steps:

[0021] S100, obtain SAR images, orbital data, and DEM files within a preset time period in a preset geographical area.

[0022] In an embodiment of the present invention, the preset geographical area may be a geographical area customized by the user. The preset time period may be a time customized by the user, as long as sufficient historical settlement data can be obtained to ensure accurate prediction. The SAR image, orbit data, and DEM file can be obtained through existing methods.

[0023] S200. Based on the SAR image, the orbit data, and the DEM file, obtain n chronological settlement data sets arranged in chronological order, and each chronological settlement data set includes the historical settlement values of m prediction points.

[0024] In an embodiment of the present invention, the SAR image is processed by integrating software packages such as ISCE and MINTPY in the Linux environment for SAR image interferometry processing and time series deformation estimation to obtain n chronological settlement data sets arranged in chronological order. Specifically, it may include the following steps:

[0025] Step 1. Based on the orbit data and the DEM file, use the ISCE integration environment to perform interferometry processing on the SAR image. As Figure 2 shown, it mainly includes steps such as baseline estimation, registration, resampling, small baseline pair formation, and generation of interferograms. Specifically, it includes the following steps:

[0026] (1) Select a suitable master image from the SAR images in SLC format;

[0027] (2) Use the obtained orbit data to register the master image and the slave image to a unified coordinate system. The slave image is the image other than the master image in the SAR images. Those skilled in the art know that any registration method for registering the master and slave images to a unified coordinate system belongs to the protection scope of the present invention.

[0028] (3) After registration, combine the external DEM file to simulate the master image intensity map and the inverted terrain phase, and perform differential interferometry on the slave image and the master image respectively to generate a set of single master image interferograms. Those skilled in the art know that any method for implementing a set of single master image interferograms belongs to the protection scope of the present invention.

[0029] (4) Screen small baseline pairs by setting the temporal baseline, spatial baseline, and coherence threshold, generate a combination of small baseline pairs, and perform differential processing on the interferogram pairs corresponding to each small baseline in the generated combination of small baseline pairs to generate a set of small baseline interferograms.

[0030] For any pixel x(r, c) in the interferogram in the set of small baseline interferograms, where r and c are the coordinates of the point in the azimuth-slant range coordinate system respectively, the phase composition of this point can be expressed as:

[0031]

[0032] Among them, is the phase contribution of surface deformation, is the phase contribution of orbital error, is the phase contribution of DEM error, is the phase contribution of atmospheric delay error, is the phase contribution of components such as background scattering, thermal noise, and decorrelation noise.

[0033] Step 2: Use MintPy for time series processing. As Figure 3 shown, it mainly includes the following steps:

[0034] 1) Process the S1A ascending orbit interferograms in the small baseline interferogram set obtained by using the minimum spanning tree and spatio-temporal baseline threshold, etc., to obtain an interferogram network.

[0035] 2) Perform phase unwrapping and error removal on the obtained interferogram network. For example, the minimum cost flow can be used to perform phase unwrapping on the interferogram network, and then phase error correction can be performed using phase closure and bridging methods.

[0036] 3) Invert the interferogram network after Step 2). In a schematic embodiment, for example, SBAS and weighted least squares can be used to invert the interferogram network.

[0037] 4) Perform atmospheric layering and ionospheric delay correction on the inverted interferogram network. In a schematic embodiment, for example, the integrated PyAPS open source software package can be used to perform atmospheric layering and ionospheric delay correction on the inverted interferogram network based on methods such as the global atmospheric model (such as ERA-5).

[0038] 5) Remove topographic errors from the interferogram network after 4). In a schematic embodiment, for example, pixel-by-pixel geometric correction and time series deformation models can be used to remove topographic errors from the interferogram network after 4).

[0039] 6) Invert the surface deformation time series of the interferogram network after 5). In a schematic embodiment, for example, geocoding can be used to invert the surface deformation time series of the interferogram network after 5) to obtain a deformation time series, that is, a time series settlement data set arranged in chronological order.

[0040] In the embodiment of the present invention, the historical settlement value is the cumulative settlement amount.

[0041] S300: Predict the settlement value at the future time node of prediction point i based on the historical settlement value of prediction point i and the historical settlement values of prediction points that have spatial correlation with prediction point i. The value of i ranges from 1 to m.

[0042] In the embodiments of the present invention, each time-series settlement dataset is embodied in matrix form. Having a spatial correlation with the prediction point i means the prediction points corresponding to the units connected to the unit where the prediction point i is located. In a schematic embodiment, the historical settlement values of the prediction point i and the historical settlement values of the prediction points having a spatial correlation with the prediction point i are obtained by using a set convolution kernel. In this way, the predicted settlement value of each prediction point takes into account the neighborhood spatial relationship, making the prediction result more accurate. In the examples of the present invention, the appropriate convolution kernel size can be determined by continuous training and testing and comparing the actual prediction effects, and different sizes of convolution kernels can be selected as needed.

[0043] Further, in the embodiments of the present invention, a set spatio-temporal sequence prediction model can be used to predict the settlement value of the future time node of the prediction point i. Preferably, the spatio-temporal sequence prediction model can be a ConvLSTM model. Further, in the embodiments of the present invention, S300 further includes:

[0044] S310, dividing the n time-series settlement datasets into n1 training sets and n2 test sets, dividing the n1 training sets into k1 training groups according to a set time series length L, and dividing the n2 test sets into k2 test groups according to the set time series length L.

[0045] The obtained time-series settlement dataset is converted into a tensor according to the date, which is convenient for convolution operation. The dataset is divided into a training set and a test set according to a certain ratio according to the time series. For example, n1:n2 = 9:1. The time series length L can be custom-set and can be an empirical value. For example, L = 12, which means using the ground settlement data of the previous 12 days to predict the ground settlement data of the 13th day. The historical settlement data in the training set is arranged in the format of [123456789101112], [2345678910111213]... to obtain k1 training groups. Similarly, the historical settlement data in the test set is arranged in the format of [123456789101112], [2345678910111213]... to obtain k2 test groups.

[0046] S320, constructing a ConvLSTM model and setting hyperparameters, and training using the k1 training groups to obtain a trained ConvLSTM model.

[0047] During the training process, the k1 training groups are sequentially input into the constructed ConvLSTM model. The model extracts data through convolution operation according to the set convolution kernel, and inputs, updates the state, forgets, and outputs the obtained convolution result according to the gate structure satisfying the following conditions:

[0048]

[0049]

[0050]

[0051]

[0052]

[0053] Among them, i is the input gate, f is the forget gate, C is the cell state, o is the output gate, and X t represents the input at the t-th moment in the time series, and H t represents the hidden state output of the corresponding cell, W is the weight coefficient matrix, b is the bias term, σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function, o represents the Hadamard product, and * represents convolution; X, C, H, i, f, and o are all three-dimensional tensors.

[0054] Hyperparameters may include the number of network layers, the number of neurons in each network layer, the size of the convolution kernel, the learning rate (learning_rate) of the entire network, the weight decay (weight_decay) of the entire network, etc.

[0055] Such as Figure 4 shown, the implementation principle of the above conditions (1) to (5) is: First, three values are input, one is the current input X t , the second is the cell state value C t-1 at the previous moment, and the last is the cell output h t-1 at the previous moment. The final outputs are the cell state value C t and the cell output H t at this moment. Gating calculations are performed separately in the cell. First, through weight calculation and the sigmoid activation function (σ), three 0-1 values are calculated, which are the gate values of the input gate, forget gate, and output gate. Then, the input value X and the output H of the previous cell are used for weight calculation, and the current input state is obtained through the tanh activation function. After integration calculation, the forget gate multiplies the state at the previous moment, and the input gate multiplies the current input state to obtain the final state C. The output gate is only related to the output, and the final output is the output gate multiplied by tanh(C).

[0056] The specific process of convolution calculation can be expressed as, for example, the convolution kernel is a P*P matrix, and the input source data of size Q*Q is scanned sequentially and the inner product is performed to obtain the output. Each time the convolution kernel moves one position, a convolution value is obtained in the output layer. After the scanning is completed, the entire convolution result is obtained, and the final output data size is (Q - P + 1)*(Q - P + 1).

[0057] In the embodiment of the present invention, the ConvLSTM model has multiple hidden layers. To avoid overfitting, when inputting data, normalization processing is performed to scale the original data proportionally so that all settlement values in each dataset are between 0 and 1.

[0058] S330, evaluate the trained ConvLSTM model, and determine the model whose evaluation result meets the first preset condition as the ConvLSTM model to be tested.

[0059] In S330, the trained ConvLSTM model can be evaluated based on the mean absolute error MAE, the mean square error MSE, and the root mean square error RMSE;

[0060] Among them, Among them, yf (i) is the predicted settlement value of the i-th prediction point in the training set corresponding to the time node of the j-th training group. The training set corresponding to the time node of the j-th training group is the training set corresponding to the time node predicted by the j-th training group. For example, if the j-th training group is [2345678910111213], then the corresponding training set is the training set with the time node of 14; y (i) is the actual settlement value of the i-th prediction point in the training set corresponding to the time node of the j-th training group.

[0061] If the MAE, MSE, and RMSE are all less than the first preset threshold, that is, the evaluation result meets the first preset condition, it means that the prediction result is accurate, and the corresponding trained ConvLSTM model is determined as the ConvLSTM model to be tested. The first preset threshold can be determined based on the data accuracy.

[0062] S340, input k2 test groups into the ConvLSTM model to be tested to obtain the test result.

[0063] Input the k2 test groups into the ConvLSTM model to be tested in sequence, and the corresponding prediction results will be obtained, that is, each prediction point will obtain the corresponding predicted settlement value.

[0064] In S350, compare the test result with the reference result corresponding to the test set. If the comparison result meets the second preset condition, determine the to-be-tested ConvLSTM model as the target ConvLSTM model.

[0065] In S350, the reference result is the corresponding historical settlement value in the test set. The test result and the reference result can be evaluated based on the mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE) in S330. To avoid redundancy, the detailed introduction is omitted here. If the MAE, MSE, and RMSE are all less than the second preset threshold, that is, the evaluation result meets the second preset condition, it means the prediction result is accurate, and then determine the to-be-tested ConvLSTM model as the target ConvLSTM model. The second preset threshold can be determined based on the data accuracy. In one example, it can be the same as the first preset threshold, and in another example, it can be different from the first preset threshold. If the evaluation result meets the second preset condition, the determined target ConvLSTM model can be visually output; if the evaluation result does not meet the second preset condition, it can be adjusted, for example, by adjusting the convolution kernel size, and then trained again.

[0066] In an illustrative embodiment of the present invention, as shown in FIG. 5(a), the convolution kernel of the obtained target ConvLSTM model can be a 3*3 matrix, that is, the convolution kernel size that can ensure the prediction result meets the preset accuracy is a 3*3 matrix. That is to say, there are 8 prediction points that have spatial correlation with each prediction point. In specific applications, use the 3*3 matrix to sequentially scan the input source data, for example, the 9*9 size source data shown in FIG. 5(b), and perform inner product to obtain the output. Each time the convolution kernel moves one position, a convolution value will be obtained in the output layer. After the scanning is completed, the entire convolution result is obtained, and the final output data size is 7*7, as shown in FIG. 5(c).

[0067] S360, use the target ConvLSTM model to predict the settlement value of the future time node of prediction point i.

[0068] In this step, for example, if it is necessary to predict the settlement value of the 13th day of prediction point i, the time series data set of the previous 12 days can be input into the target ConvLSTM model.

[0069] The embodiment of the present application also provides a non-transitory computer-readable storage medium, which can be set in an electronic device to store at least one instruction or at least one program related to a method in the method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the method provided in the above embodiment.

[0070] An embodiment of the present application further provides an electronic device, including a processor and the aforementioned non-transitory computer-readable storage medium.

[0071] An embodiment of the present application further provides a computer program product, which includes program code. When the program product runs on an electronic device, the program code is used to cause the electronic device to execute the steps in the methods according to various exemplary embodiments of the present application described above in this specification.

[0072] Although some specific embodiments of the present application have been described in detail by way of examples, those skilled in the art should understand that the above examples are for illustrative purposes only and not for limiting the scope of the present application. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the present application. The scope of the present application is defined by the appended claims.

Claims

1. A ground settlement prediction method based on spatio-temporal sequence data, characterized in that, The method includes the following steps: S100. Obtain SAR images, orbital data, and DEM files within a preset time period in a preset geographical area; S200. Based on the SAR images, the orbital data, and the DEM files, obtain n time-series settlement data sets arranged in chronological order, and each time-series settlement data set includes historical settlement values of m prediction points; S300. Predict the settlement value of the future time node of prediction point i based on the historical settlement value of prediction point i and the historical settlement values of prediction points that have a spatial correlation with prediction point i, where the value of i ranges from 1 to m; Use a ConvLSTM model to predict the settlement value of the future time node of prediction point i; S300 further includes: S310. Divide the n time-series settlement data sets into n1 training sets and n2 test sets, divide the n1 training sets into k1 training groups according to a set time-series length L, and divide the n2 test sets into k2 test groups according to the set time-series length L; S320. Construct a ConvLSTM model and set hyperparameters, and use the k1 training groups to train the constructed ConvLSTM model to obtain a trained ConvLSTM model; S330. Evaluate the trained ConvLSTM model, and determine the model whose evaluation result meets the first preset condition as the ConvLSTM model to be tested; S340. Input the k2 test groups into the ConvLSTM model to be tested to obtain test results; S350. Compare the test results with the reference results corresponding to the k2 test groups. If the comparison result meets the second preset condition, determine the ConvLSTM model to be tested as the target ConvLSTM model; S360. Use the target ConvLSTM model to predict the settlement value of the future time node of prediction point i.

2. The method according to claim 1, wherein Obtain the historical settlement value of prediction point i and the historical settlement values of prediction points that have a spatial correlation with prediction point i by using a set convolution kernel.

3. The method according to claim 1, wherein In S320, extract data through convolution operations according to the set convolution kernel, and input, update the state, forget, and output the obtained convolution results according to a gate structure that meets the following conditions: Among them, i is the input gate, f is the forget gate, C is the cell state, is the output gate, X t represents the input at the t-th moment in the time series, H t represents the hidden state output of the corresponding cell, W is the weight coefficient matrix, b is the bias term, σ is the sigmoid activation function, and tanh is the hyperbolic tangent activation function, represents the Hadamard product, and * represents convolution.

4. The method according to claim 1, wherein In S330, evaluate the trained ConvLSTM model based on the mean absolute error MAE, the mean square error MSE, and the root mean square error RMSE; Among them, Among them, yf (i) is the predicted settlement value of the i-th prediction point in the training set corresponding to the time node corresponding to the j-th training group, and y (i) is the actual settlement value of the i-th prediction point in the training set corresponding to the time node corresponding to the j-th training group; If the MAE, MSE, and RMSE are all less than a preset threshold, determine the corresponding trained ConvLSTM model as the ConvLSTM model to be tested.

5. The method according to claim 1, wherein The ConvLSTM model has multiple hidden layers.

6. The method according to claim 1, characterized in that, The convolution kernel of the target ConvLSTM model is a 3*3 matrix.

7. A non-transitory computer-readable storage medium, in which at least one instruction or at least one program segment is stored, and the at least one instruction or the at least one program segment is loaded and executed by a processor to implement the method according to any one of claims 1 to 6.

8. An electronic device, characterized in that, It includes a processor and the non-transitory computer-readable storage medium according to claim 7.

Images