Mining area ground surface settlement prediction method based on VMD-SSA-LSTM

Abstract

The invention belongs to the technical field of address disaster prediction, discloses and relates to a mining area ground surface settlement prediction method based on VMD-SSA-LSTM, and aims to improve the precision and reliability of ground surface settlement prediction. Comprising the following steps: acquiring multi-temporal ground surface settlement data of a mining area, and preprocessing the data; decomposing the preprocessed data by adopting variational mode decomposition, and extracting intrinsic mode functions of different frequency components; performing singular spectrum analysis on the main IMF; constructing and training a long short-term memory network model, inputting a time sequence after SSA reconstruction, outputting a future ground surface settlement prediction value, and optimizing model parameters by using a cross validation method; and finally, ground surface settlement prediction is carried out through the trained model. According to the method, multi-source data and advanced signal processing and machine learning technologies are comprehensively utilized, the efficient and accurate mining area ground surface settlement prediction method is provided, and the method can be applied to early warning and prevention of mining area geological disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster prediction, and in particular to a method for predicting surface subsidence in a mining area, which predicts surface subsidence based on a small baseline aggregate synthetic aperture radar interferometry (SBAS-InSAR) and a variational mode decomposition (VMD)-singular spectrum analysis (SSA)-long short-term memory network (LSTM) model. Background Art

[0002] Surface subsidence is a phenomenon in which the surface height decreases due to natural or human activities, which often poses a serious threat to infrastructure, buildings and the ecological environment. Traditional methods for monitoring surface subsidence include ground measurement and GPS monitoring, but these methods have shortcomings such as low spatial resolution, limited coverage, and susceptibility to weather. Small baseline integrated synthetic aperture radar interferometry (SBAS-InSAR) technology can provide large-scale, high-precision surface subsidence monitoring data. However, SBAS-InSAR technology mainly provides historical monitoring data and its prediction ability is limited.

[0003] In recent years, machine learning technology, especially long short-term memory network (LSTM), has shown significant advantages in processing time series data. However, a single LSTM model is difficult to fully utilize the complex characteristics of surface subsidence data. Therefore, it is necessary to invent a new method for predicting surface subsidence in mining areas to better extract effective information from the data and improve the prediction accuracy. Summary of the invention

[0004] In order to solve the problem that surface subsidence is difficult to accurately predict in the prior art, the present invention proposes a mining area surface subsidence prediction method based on VMD-SSA-LSTM, which can achieve more accurate and efficient subsidence prediction based on high-precision monitoring of surface subsidence.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for predicting surface subsidence in mining areas based on VMD-SSA-LSTM, comprising the following steps:

[0006] Step 1: Use synthetic aperture radar to obtain surface data, and preprocess the data based on multi-temporal synthetic aperture radar interferometry to obtain surface monitoring data of the study area;

[0007] Step 2: Perform variational mode decomposition on the preprocessed surface monitoring data, decompose it into multiple intrinsic mode functions, and select the main intrinsic mode functions for subsequent analysis;

[0008] Step 3: Perform singular spectrum analysis on the selected intrinsic mode function to extract the main trend component and periodic component; then reconstruct the extracted main trend component and periodic component to obtain a smooth time series;

[0009] Step 4: Construct an LSTM model, with the input being the reconstructed time series and the output being the predicted value of surface subsidence at future moments; optimize the LSTM model parameters through training data, and train to obtain the final LSTM prediction model;

[0010] Step 5: Obtain the surface monitoring data of the study area, process it through steps 2 to 3 above, and use the trained LSTM model to predict future surface subsidence.

[0011] In the step 1, the pretreatment comprises the following steps:

[0012] Conduct baseline estimates of surface data;

[0013] The primary image and the secondary image are differentiated based on the baseline estimation result to generate an interference map;

[0014] Based on the interference pattern, the image is phase unwrapped;

[0015] Based on the phase unwrapping results, the image is geocoded.

[0016] In the step 2, when performing variational mode decomposition on the preprocessed surface settlement data, it also includes the step of adjusting VMD parameters to optimize the decomposition result of the intrinsic mode function.

[0017] The specific steps of step 3 are:

[0018] Time series embedding: embed the original time series into a high-dimensional space to form a trajectory matrix;

[0019] Singular value decomposition: Perform singular value decomposition on the trajectory matrix to obtain eigenvectors and singular values;

[0020] Select components: Select the characteristic components corresponding to the larger singular values ​​according to the size of the singular values;

[0021] Reconstruct components: reconstruct the selected eigenvectors and corresponding singular values ​​back into the time series to obtain the main trend components and periodic components;

[0022] Combined components: The main trend component and the cycle component are combined to form a smooth time series.

[0023] In step 4, the constructed LSTM model is optimized in parameters by a cross-validation method.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] This paper proposes a method for predicting surface subsidence in mining areas based on VMD-SSA-LSTM, which combines high-precision SBAS-InSAR data, multi-level VMD and SSA signal processing technology, and powerful LSTM time series analysis capabilities to accurately extract and predict surface subsidence trends. Through automated data processing and cross-validated parameter optimization, the prediction accuracy and efficiency are improved, providing important technical support for geological disaster warning and prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flowchart of a method for predicting surface subsidence in a mining area based on VMD-SSA-LSTM provided in an embodiment of the present invention;

[0027] Figure 2 is a flow chart of the SBAS-InSAR algorithm used in an embodiment of the present invention;

[0028] Figure 3 1 is a basic structural diagram of the LSTM model used in the embodiment of the present invention;

[0029] Figure 4 is the overall average deformation rate diagram obtained in the embodiment of the present invention;

[0030] Figure 5 This is a diagram of the process of layer-by-layer decomposition of data by VMD in an embodiment of the present invention;

[0031] Figure 6 This is a comparison chart of prediction errors of the VMD-SSA-LSTM model used in the embodiment of the present invention and the comparison model on the test set;

[0032] Figure 7 This is a comparison chart of the prediction results of the VMD-SSA-LSTM model used in the embodiment of the present invention and the comparison model. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0034] like Figure 1 As shown, the embodiment of the present invention provides a method for predicting surface subsidence in a mining area based on VMD-SSA-LSTM, comprising the following steps:

[0035] Step 1: Use synthetic aperture radar to obtain surface data, and preprocess the data based on multi-temporal synthetic aperture radar interferometry (SBAS-InSAR) to obtain surface subsidence monitoring data in the study area.

[0036] In this embodiment, the SBAS-InSAR method is a technology that uses synthetic aperture radar (SAR) data to monitor surface deformation. Its core principle is to select SAR image pairs with short temporal and spatial baselines (i.e., small temporal and spatial differences) and perform differential interferometry processing to obtain surface deformation information.

[0037] Assume that there are two SAR images, acquired at time t A and t B , for a certain pixel (x, y) in the image, its differential interference phase δΦ(x, y) can be expressed as:

[0038] δΦ(x,y)=Φ(t B ,x,y)―Φ(t A ,x,y); (1)

[0039] Among them, Φ(t B ,x,y) and Φ(t A ,x,y) are t B and t A The interference phase of the pixel at the moment.

[0040] According to the basic principle of radar interferometry, the relationship between the surface deformation d(x, y) and the differential interferometric phase δΦ(x, y) can be expressed as:

[0041]

[0042] Where λ is the radar wavelength and θ is the incident angle of the radar beam. Therefore, the surface deformation information can be obtained based on the interference phase.

[0043] Specifically, Figure 2 As shown, in step 1, the pretreatment includes the following steps:

[0044] Conduct baseline estimates of surface data;

[0045] The primary image and the secondary image are differentiated based on the baseline estimation result to generate an interference map;

[0046] Based on the interference pattern, the image is phase unwrapped;

[0047] Based on the phase unwrapping results, the image is geocoded.

[0048] Step 2: Perform variational mode decomposition (VMD) on the preprocessed surface monitoring data to decompose it into multiple intrinsic mode functions (IMFs), and select the main intrinsic mode functions for subsequent analysis.

[0049] Specifically, the main intrinsic mode functions can be screened out by analyzing the frequency characteristics and time series characteristics of multiple intrinsic mode functions obtained by decomposition. The main intrinsic mode functions should contain the most significant frequency components in the data, and spectrum analysis can be used to determine which intrinsic mode functions contain the most important frequency information; in addition, the main intrinsic mode functions should be able to retain the main time series characteristics in the original data, such as trend.

[0050] In the step 2, when performing variational mode decomposition on the preprocessed surface settlement data, it also includes the step of adjusting VMD parameters to optimize the decomposition result of the intrinsic mode function.

[0051] Step 3: Perform singular spectrum analysis (SSA) on the selected main intrinsic mode functions to extract the main trend components and periodic components; then reconstruct the extracted main trend components and periodic components to obtain a smooth time series.

[0052] The specific steps of step 3 are:

[0053] Time series embedding: embed the original time series corresponding to the selected intrinsic mode function into a high-dimensional space to form a trajectory matrix;

[0054] Singular value decomposition: Perform singular value decomposition on the trajectory matrix to obtain eigenvectors and singular values;

[0055] Select components: Select the characteristic components corresponding to the larger singular values ​​according to the size of the singular values, so that the main trend components and periodic components can be selected;

[0056] Reconstruct components: reconstruct the selected eigenvectors and corresponding singular values ​​back into the time series to obtain the main trend components and periodic components;

[0057] Combined components: The main trend component and the cycle component are combined to form a smooth time series.

[0058] Step 4: Construct an LSTM model, with the input being the reconstructed time series and the output being the predicted value of surface subsidence at future moments; optimize the LSTM model parameters through training data and train to obtain the final LSTM prediction model.

[0059] In step 4, the constructed LSTM model is optimized in parameters by a cross-validation method.

[0060] like Figure 3As shown in Figure 1, LSTM (Long Short-Term Memory) is a special recurrent neural network (RNN) that solves the gradient vanishing and gradient exploding problems in traditional RNN by introducing gating mechanisms such as forget gate, input gate, and output gate.

[0061] Input Gate i t Controls the extent to which new information enters the cell state:

[0062] i t =σ(W i ·[h t―1 ,x t ]+b i ); (3)

[0063] Among them, x t is the current input; h t-1 is the hidden state of the previous time step; W i is the weight matrix of the input gate; b i is the bias vector of the input gate; σ is the sigmoid activation function, which is used to compress the output value to between [0,1].

[0064] Forget Gate t Determines the extent to which old information in the cell state is forgotten:

[0065] f t =σ(W f ·[h t―1 ,x t ]+b f ); (4)

[0066] Candidate memory cells Generate new candidate memory values ​​and process them through the tanh activation function to decide which new information is added to the cell state:

[0067]

[0068] The cell state is the core of LSTM and represents the storage method of cell memory. The cell state is updated as follows:

[0069]

[0070] Among them, the forget gate f t Determined old memory C t-1 How much is retained; input gate i t Determined new memory C t How many candidate memory cells are added

[0071] Specifically, the training data may be historical data acquired based on a multi-temporal synthetic aperture radar interferometry method (SBAS-InSAR).

[0072] Step 5: Obtain the surface monitoring data of the study area, process it through steps 2 to 3 above, and use the trained LSTM model to predict future surface subsidence.

[0073] Prediction and verification: Use the trained LSTM model to predict the surface subsidence at future times and obtain the prediction results. Compare the prediction results with the actual observation data to evaluate the prediction accuracy and reliability of the model to ensure that the model can accurately predict future surface subsidence.

[0074] like Figure 4 As shown in FIG. 1 , it is a graph of the overall average deformation rate of the surface data obtained in an embodiment of the present invention; Figure 5 As shown in FIG. 1 , it is a schematic diagram of the IMF obtained by performing WMD layer-by-layer decomposition of surface data.

[0075] like Figure 6 As shown, it is a comparison diagram of the prediction errors of the VMD-SSA-LSTM model and the comparison models (LSTM and VMD-LSTM) used in the embodiment of the present invention on the test set; wherein, LSTM is to use the direct LSTM model to directly predict the SBAS-InSAR data, and VMD-LSTM is to predict the SBAS-InSAR data through the LSTM model after performing variational mode decomposition; in LSTM and VMD-LSTM, the data input to the LSTM model is the time pixel settlement value of the settlement data pixel monitored by INSAR, which is extracted from the settlement image obtained by monitoring based on INSAR technology. Figure 6 It shows that the technical solution of the present invention has a smaller prediction error.

[0076] Figure 7 This is a comparison chart of the prediction results of the VMD-SSA-LSTM model used in the embodiment of the present invention and the comparison model. Figure 7 This shows that the prediction result of the present invention is closer to the true value, so the present invention has higher prediction accuracy.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting surface subsidence in mining areas based on VMD-SSA-LSTM, characterized in that: The following steps are involved: Step 1: Use synthetic aperture radar to obtain surface data, and preprocess the data based on multi-temporal synthetic aperture radar interferometry to obtain surface monitoring data of the study area; Step 2: Perform variational mode decomposition on the preprocessed surface monitoring data, decompose it into multiple intrinsic mode functions, and select the main intrinsic mode functions for subsequent analysis; Step 3: Perform singular spectrum analysis on the selected intrinsic mode function to extract the main trend component and periodic component; then reconstruct the extracted main trend component and periodic component to obtain a smooth time series; Step 4: Construct an LSTM model, with the input being the reconstructed time series and the output being the predicted value of surface subsidence at future moments; optimize the LSTM model parameters through training data, and train to obtain the final LSTM prediction model; Step 5: Obtain the surface monitoring data of the study area, process it through steps 2 to 3 above, and use the trained LSTM model to predict future surface subsidence.

2. The method for predicting surface subsidence in mining areas based on VMD-SSA-LSTM according to claim 1, characterized in that: In the step 1, the pretreatment comprises the following steps: Conduct baseline estimates of surface data; The primary image and the secondary image are differentiated based on the baseline estimation result to generate an interference map; Based on the interference pattern, the image is phase unwrapped; Based on the phase unwrapping results, the image is geocoded.

3. The method for predicting surface subsidence in mining areas based on VMD-SSA-LSTM according to claim 1, characterized in that: In the step 2, when performing variational mode decomposition on the preprocessed surface settlement data, it also includes the step of adjusting VMD parameters to optimize the decomposition result of the intrinsic mode function.

4. The method for predicting surface subsidence in mining areas based on VMD-SSA-LSTM according to claim 1, characterized in that: The specific steps of step 3 are: Time series embedding: embed the original time series into a high-dimensional space to form a trajectory matrix; Singular value decomposition: Perform singular value decomposition on the trajectory matrix to obtain eigenvectors and singular values; Select components: Select the characteristic components corresponding to the larger singular values ​​according to the size of the singular values; Reconstruct components: reconstruct the selected eigenvectors and corresponding singular values ​​back into the time series to obtain the main trend components and periodic components; Combined components: The main trend component and the cycle component are combined to form a smooth time series.

5. The method for predicting surface subsidence in mining areas based on VMD-SSA-LSTM according to claim 1, characterized in that: In step 4, the constructed LSTM model is optimized in parameters by a cross-validation method.

6. The method for predicting surface subsidence in mining areas based on VMD-SSA-LSTM according to claim 1, characterized in that: In step 4, the training data is historical data obtained based on multi-temporal synthetic aperture radar interferometry.

Images