A long-time series prediction method, system, medium and device for surface subsidence in mining areas

Abstract

The present disclosure provides a long-term surface subsidence prediction method, system, medium and device for mining areas, which relates to the technical field of InSAR and geological disaster prevention and control. The initial surface subsidence rate of the mining area is obtained, and after preprocessing the initial surface subsidence rate, long-term InSAR subsidence data at equal time intervals is obtained; the cumulative surface subsidence amount of the mining area is calculated by using the InSAR subsidence data; the mining area is partitioned according to the cumulative subsidence amount by using threshold division, and according to the mining area partition result, the spatial neighborhood information of point targets in different mining area partitions is obtained; the spatial neighborhood information is input into an improved Informer model to learn the long-term non-linear characteristics of local positions in different partitions, and the temporal subsidence prediction value of each pixel is output to realize the long-term prediction of the subsidence information of the mining area; the present disclosure reduces the time complexity and memory occupancy problems of the original Transformer under long sequence input.

Description

Technical Field

[0001] The present disclosure relates to the technical field of InSAR and geological disaster prevention and control, and particularly relates to a long-time series prediction method, system, medium and device for surface subsidence in mining areas. Background Art

[0002] The statements in this part merely provide background technical information related to the present disclosure, and do not necessarily constitute prior art.

[0003] Coal is the main basic energy source and important raw material in the world, and has great strategic significance for the development of an economy. However, long-term coal mining has also brought serious negative ecological environment effects to coal resource areas, such as geological disasters induced by coal mining, such as surface subsidence, landslides, and soil erosion, causing damage to structures such as roads, bridges, farmland, and houses, and seriously threatening the safety of local residents' production and life. Therefore, carrying out long-time series prediction of surface subsidence in mining areas is of great significance for geological disaster prevention and control, restoration of mined-out areas in mining areas, and ensuring the safety of the lives and property of surrounding residents.

[0004] InSAR uses the phase information of two or more radar images covering the same area to quickly extract the elevation information and subsidence information in the satellite line-of-sight direction of a large area of the ground, and has advantages such as low cost, high spatio-temporal resolution, and no need to rely on control points, which are incomparable to traditional geodetic techniques such as leveling, three-dimensional laser scanning, and global positioning system.

[0005] In existing research, InSAR is often used as a data acquisition means for surface subsidence monitoring, lacking an integrated monitoring-prediction model. In fact, in the prevention and control of mine geological disasters, monitoring and prediction are inseparable. Only by integrating mine subsidence monitoring and prediction can the surface be monitored to timely understand the deformation situation and future deformation characteristics, providing a reasonable basis for mine management decisions. With the rapid development of artificial intelligence, deep learning methods have been widely applied in fields such as computer vision. A deep learning network is a multi-layer perceptron containing multiple hidden layers. By learning from samples, a deep non-linear network structure is obtained to realize the approximation of complex functions. For complex data, deep learning can extract the essential features of the data by combining low-level features and abstracting high-level features, providing a new solution idea for predicting surface deformation in mining areas.

[0006] Most of the deep learning methods widely used for time series prediction are based on recurrent structure networks, such as RNN (Recurrent Neural Network) and LSTM (Short-Term Memory). Although such models can better solve the sequence prediction problem compared with other deterministic and traditional statistical models, since they extract time series information sequentially and continuously pass it backward, it is difficult or impossible to train the model. At the same time, during the sequential calculation process, once information is lost. Although the gated recurrent unit has the ability to alleviate the long-term dependence to a certain extent, it is difficult to completely handle the long-term dependence phenomenon. Summary of the Invention

[0007] To solve the above problems, the present disclosure proposes a long-time series prediction method, system, medium and device for surface subsidence in mining areas. Based on the improved Informer and InSAR, the long-time series prediction of the subsidence change trend in the mining area is carried out, and the time series prediction values of more than 12 timestamps of subsidence points in the study area are obtained, so as to realize the long-time series prediction of the subsidence information in the mining area.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions:

[0009] A long-time series prediction method for surface subsidence in mining areas, comprising:

[0010] Obtain the initial surface subsidence rate in the mining area, and preprocess the initial surface subsidence rate, including denoising and time series difference, to fill in missing values and improve the quality of the data, so as to obtain long-time series InSAR subsidence data with equal time intervals and significant time series change characteristics;

[0011] Calculate the cumulative surface subsidence amount in the mining area by using the InSAR subsidence data; divide the mining area into zones according to the threshold based on the cumulative subsidence amount, and the zoning basis includes the magnitude of the subsidence amount and the change of the subsidence rate; then in each mining area zone, extract the spatial neighborhood information of the point target;

[0012] Input the spatial neighborhood information into the improved Informer model to learn the long-time series non-linear characteristics of the subsidence points in different zones, and output the time series subsidence prediction values of each subsidence point, so as to realize the long-time series prediction of the subsidence information in the mining area.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] A long-time series prediction system for surface subsidence in mining areas, comprising:

[0015] A data acquisition module that acquires the initial surface settlement rate of the mining area, including denoising and time series difference, to fill in missing values and improve data quality, thereby obtaining long-time series InSAR settlement data with equal time intervals and significant time series change characteristics;

[0016] Calculate the cumulative surface settlement amount of the mining area using the InSAR settlement data; divide the mining area according to the threshold using the cumulative settlement amount, and the partitioning basis includes the size of the settlement amount and the change in the settlement rate; then, in each mining area partition, extract the spatial neighborhood information of the point target;

[0017] A prediction module for inputting the spatial neighborhood information into an improved Informer model to learn the long-time series non-linear characteristics of the settlement points in different partitions, and outputting the time series settlement prediction values of the settlement points to achieve long-time series prediction of the settlement information in the mining area.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions:

[0019] A non-transitory computer-readable storage medium for storing computer instructions, which when executed by a processor, implement the long-time series mining area surface settlement prediction method described above.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] An electronic device includes: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory to enable the electronic device to execute the long-time series mining area surface settlement prediction method described above.

[0022] Compared with the prior art, the beneficial effects of the present disclosure are:

[0023] The present disclosure proposes a long-time series mining area surface settlement prediction method, specifically a long-time series mining area surface settlement prediction method based on an improved Informer and InSAR. By extracting the settlement prediction values of more than 12 time stamps from the existing InSAR mining area surface settlement data in the study area, and using the iTranformer Block and attention distillation mechanism proposed by the improved Informer model to reduce the time complexity and memory occupancy problems of the original Transformer under long sequence input. Finally, long-time series prediction of mining area settlement can be achieved, providing auxiliary decision-making and technical support for mining area settlement disasters. Description of the Drawings

[0024] The accompanying drawings forming a part of this disclosure are used to provide a further understanding of the disclosure. The schematic embodiments and descriptions thereof of the disclosure are used to explain the disclosure and do not constitute an improper limitation of the disclosure.

[0025] Figure 1 It is a schematic diagram of the overall process of the long-term mining area surface subsidence prediction method according to the embodiment of the present disclosure;

[0026] Figure 2 It is a schematic diagram of the mining area zoning result according to the embodiment of the present disclosure;

[0027] Figure 3 It is a schematic diagram of the improved Informer model structure according to the embodiment of the present disclosure;

[0028] Figure 4 It is a schematic diagram of the iTranformer block structure according to the embodiment of the present disclosure. Detailed implementation manners

[0029] The present disclosure will be further described below in conjunction with the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present disclosure belongs.

[0031] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0032] Embodiment 1

[0033] In one embodiment of the present disclosure, a long-term mining area surface subsidence prediction method is provided, including the following steps:

[0034] Step 1: Obtain the initial surface subsidence rate of the mining area. After preprocessing the initial surface subsidence rate, including denoising and time series difference to fill in missing values and improve the quality of the data, long-term InSAR subsidence data with equal time intervals and significant time series change characteristics can be obtained;

[0035] Step 2: Calculate the cumulative surface settlement of the mining area using the InSAR settlement data; divide the mining area based on the change trends of the normalized cumulative settlement and settlement rate, and obtain the spatial neighborhood information of point targets in different sub - areas of the mining area according to the sub - area results; construct a training data set and a test data set;

[0036] Step 3: Input the spatial neighborhood information into the improved Informer model to learn the long - time - series non - linear features of settlement points in different sub - areas, and output the time - series settlement prediction values of settlement points, so as to realize the long - time - series prediction of settlement information in the mining area.

[0037] As an embodiment, the long - time - series surface settlement prediction method for a mining area based on Infomer and InSAR can realize the long - time - series prediction of mining area settlement, specifically including:

[0038] Step 1: Obtain the initial surface settlement rate of the mining area through InSAR time - series processing technology, filter and interpolate the initial surface settlement rate to obtain InSAR settlement rate data at equal time intervals;

[0039] Step 2: Calculate the cumulative surface settlement of the mining area using the InSAR settlement rate data;

[0040] After normalizing the obtained cumulative settlement results, divide the mining area by combining the normalized cumulative settlement results and the change trend of the settlement rate, and process the spatial neighborhood information of point targets in different sub - areas according to the sub - area results;

[0041] Construct a training data set and a test data set; train different models in each sub - area respectively to learn the long - time - series non - linear features of local positions in different sub - areas, and evaluate the network performance in the test data at the same time;

[0042] Step 3: Use the trained improved Informer model for early prediction to obtain the settlement prediction values of each settlement point exceeding 12 time stamps, so as to realize the long - time - series prediction of settlement information in the mining area.

[0043] In the above - mentioned implementation manner, the process of obtaining long - time - series InSAR settlement data at equal time intervals includes:

[0044] Obtain the initial surface settlement rate of the mining area using InSAR time - series processing technology;

[0045] Take twice the local median absolute deviation of the surface settlement rate as the threshold, and filter out abnormal data from the original surface settlement rate by setting a sliding window method to obtain a new surface settlement rate;

[0046] Perform cubic spline interpolation on the filtered ground settlement rate to solve the problem of unequal time intervals of monitoring data caused by the irregular revisit of SAR satellites. Define the time series data with equal time intervals after interpolation as X = {X1, X2,..., X m+p}.

[0047] In step 2, the construction of the training dataset and the test dataset includes:

[0048] According to the time series data X with equal time intervals, calculate the cumulative settlement data, and normalize the cumulative settlement data using the standard deviation normalization method. Combine the normalized results to conduct spatial partitioning of the study area. As Figure 2 shown, divide the study area into 4 sub-regions. The settlement characteristics of each sub-region are similar and the spatial positions are adjacent. Training in each partition helps the model to obtain more reliable results;

[0049] Extract the spatial neighborhood information of the point targets included in the time series data X and save it to the training dataset Y according to the partitioning results. Suppose there are n sample points with a time series length of m + p. Use the data on the first m time series of the sample points as the training set, denoted as where represents the partition, and the data on the remaining p time series are used as the test set, denoted as Then and can be respectively expressed as:

[0050]

[0051]

[0052] In step 3, use the trained improved Informer model for early prediction, as Figure 3As shown in the figure, the overall structure of the improved Informer model consists of two parts: an encoder and a decoder. Specifically, the encoder and the decoder contain iTranformer blocks and self-attention distillation layers. Among them, the iTranformer block mainly includes an embedding layer, a multivariate attention mechanism, a normalization layer, a feed-forward network, and a fully connected layer. Different from the traditional Transformer module, the embedding layer of the iTranformer block takes all the data on the time series of a sample point as a group and inputs it into the network, and the time series can be regarded as an independent process. The multivariate attention mechanism obtains Q (query vector), K (key vector), and V (value vector) through linear projection using the self-attention module, comprehensively extracts the temporal features of this sample point, and reveals the correlation between variables. The normalization layer can reduce the differences between different sample points. The feed-forward network is used to extract non-linear features. The projection layer performs a linear projection operation on the processed result to obtain the final predicted value. The self-attention distillation layer uses KL divergence to extract data with a larger information entropy, thereby reducing the dimension and the number of network parameters.

[0053] The input of the encoder is a long sequence composed of the spatial neighborhood information of each point target in the mining subsidence area during the time period with known settlement information. First, the iTranformer block is used to comprehensively extract the time series features and fully explain the correlation between variables. Then, the self-attention distillation layer is used to reduce the dimension. Finally, the output of the encoder is obtained.

[0054] The input of the decoder consists of two parts: the first part is the combination of the short sequence at the end of the time series and 0 values equal to the prediction step length. Among them, the 0 value input to the decoder is used as a placeholder for the mining subsidence prediction value. The second part is the output of the encoder. After using the iTranformer block to extract features from the first part, it is combined with the second part, and the multi-head attention module is used for deep feature extraction. Finally, the data dimension is adjusted through the fully connected layer to obtain the prediction result of the surface subsidence in the mining area. The output prediction result is backpropagated through the reverse gradient after calculating the mean square error (MSE) loss, and the network parameters are continuously optimized.

[0055] In the above embodiment, the settlement prediction includes:

[0056] Performing a partitioning operation on the obtained data according to the partitioning method to obtain verification data for different partitions;

[0057] After processing the verification data for different partitions according to the data filtering and interpolation method, the processed data is input into the trained network to obtain long-time series settlement prediction values of each settlement point exceeding 12 time stamps.

[0058] Example 2

[0059] In one embodiment of the present disclosure, a long-term mine surface subsidence prediction system is provided, including:

[0060] A data acquisition module that acquires the initial surface subsidence rate of the mine area, and after preprocessing the initial surface subsidence rate, acquires long-term InSAR subsidence data at equal time intervals;

[0061] Calculate the cumulative surface subsidence amount of the mine area using the InSAR subsidence data; divide the mine area into zones according to the cumulative subsidence amount using threshold partitioning, and according to the mine area zoning result, obtain the spatial neighborhood information of point targets within different mine area zones;

[0062] A prediction module for inputting the spatial neighborhood information into an improved Informer model to learn the long-term non-linear features of local positions within different zones, and outputting the time-series subsidence prediction values of each subsidence point exceeding 12 time stamps, so as to realize the long-term prediction of the mine area subsidence information.

[0063] Embodiment 3

[0064] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the long-term mine surface subsidence prediction method described above is implemented.

[0065] Embodiment 4

[0066] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory, so that the electronic device executes the long-term mine surface subsidence prediction method described above.

[0067] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions for implementing the steps specified in one process or a plurality of processes and / or blocks Figure 1 in one block or a plurality of blocks Figure 1 for the functions specified in one block or a plurality of blocks.

[0069] Although the specific embodiments of the present disclosure have been described in conjunction with the accompanying drawings, they are not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that, based on the technical solutions of the present disclosure, various modifications or variations that can be made without creative efforts by those skilled in the art are still within the scope of protection of the present disclosure.

Claims

1. A method for predicting surface subsidence in a mining area over a long time series, characterized in that: include: Obtain the initial surface subsidence rate of the mining area, and after preprocessing the initial surface subsidence rate, obtain long-time series InSAR subsidence data with equal time intervals; Calculating the accumulated surface subsidence of the mining area using the InSAR subsidence data; According to the accumulated settlement, the mining area is divided into zones using threshold values, and according to the mining area division results, the spatial neighborhood information of point targets in different mining area divisions is obtained; The process of obtaining long-time series InSAR subsidence data with equal time intervals includes constructing a training data set and a test data set using spatial neighborhood information, specifically: calculating cumulative subsidence data based on time series data with equal time intervals, and normalizing the cumulative subsidence data using a standard deviation normalization method, and spatially partitioning the study area based on the normalized results to obtain sub-areas with similar subsidence characteristics and adjacent spatial positions; Extract the spatial neighborhood information of point targets contained in the time series data and save it into the training data set according to the partition results; The spatial neighborhood information is input into the improved Informer model to learn the long-term nonlinear characteristics of settlement points in different partitions, and the time-series settlement prediction value of the settlement point is output to achieve the long-term prediction of the settlement information of the mining area. The overall structure of the improved Informer model includes an encoder and a decoder, and the encoder and the decoder include an iTranformer block and a self-attention distillation layer, wherein the iTranformer block mainly includes an embedding layer, a multivariate attention mechanism, a normalization layer, a feedforward network and a fully connected layer. The embedding layer of the iTranformer block inputs all data on a sample point time series into the network as a group, and regards the time series as an independent process; the input of the encoder is a long sequence composed of the spatial neighborhood information of each point target in the mining subsidence area within the time period of known settlement information. First, the iTranformer block is used to comprehensively extract the time series features and fully explain the correlation between the variables. The iTranformer Block and the attention distillation mechanism proposed by the improved Informer model are used to reduce the time complexity and memory occupancy rate of the original Transformer under long sequence input, so as to realize the long time series prediction of mining area settlement.

2. A method for predicting surface subsidence in a mining area over a long time series as claimed in claim 1, characterized in that: in, The 0 value input to the decoder is used as a placeholder for the mining subsidence prediction value.

3. A long-term mining area surface subsidence prediction system, characterized in that: Executing a method for predicting surface subsidence in a mining area over a long time series as claimed in any one of claims 1 to 2 comprises: The data acquisition module obtains the initial surface settlement rate of the mining area, and obtains the long-time series InSAR settlement data with equal time intervals after preprocessing the initial surface settlement rate; Calculate the accumulated surface settlement of the mining area using the InSAR settlement data; partition the mining area using threshold value according to the accumulated settlement, and obtain spatial neighborhood information of point targets in different mining area partitions based on the mining area partition results; The prediction module is used to input the spatial neighborhood information into the improved Informer model to learn the long-term nonlinear characteristics of local positions in different partitions, output the time-series settlement prediction value of each settlement point, and realize the long-term prediction of the settlement information of the mining area.

4. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, a long-time series mining area surface subsidence prediction method as described in any one of claims 1-2 is implemented.

5. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes a long-time series mining area surface subsidence prediction method as described in any one of claims 1-2.

Images