A geological method and device for intelligent prediction of coal mining subsidence

Abstract

The present invention discloses a geological method and device for intelligent prediction of coal mining subsidence, and relates to the technical field of prevention and control of mining or geological disasters and geological environmental problems. The geological method for intelligent prediction of coal mining subsidence includes: collecting geological information, hydrological information, mining activities and other geological data that restrict surface subsidence in the mining area; using InSAR monitoring technology to obtain high-precision surface deformation time series data of the mining area; combining geological data and surface deformation data, and predicting coal mining subsidence based on a kernel algorithm. The geological method for intelligent prediction of coal mining subsidence provided by the present invention realizes intelligent prediction and analysis of surface subsidence in the mining area, improves the accuracy and efficiency of coal mining subsidence prediction, provides a scientific basis for the control and management of coal mining subsidence, helps to reduce the impact of mining activities on the environment, and ensures the safety and sustainable development of coal mining, and has good application value and promotion prospects.

Description

Technical Field

[0001] The present invention belongs to the technical field of prevention and control of mining or geological disasters and geological environmental problems, and specifically relates to a geological method and a geological device for intelligent prediction of coal mining subsidence. Background Art

[0002] During coal mining, as the mine or coal mining face advances, the coal pillars and the rock layers above the coal seams will fall, break, bend and move due to loss of support, which will eventually cause surface subsidence. Surface subsidence poses a serious threat to normal human production and life, significantly affecting the harmonious coexistence of man and nature. It often causes land resource problems such as surface morphology destruction, increased water and soil fluid and soil erosion, changes in soil moisture and nutrients, and soil quality degradation; it causes water resource damage problems such as surface water interruption and groundwater leakage; it leads to ecological and environmental problems such as plant growth and development obstruction and changes in vegetation types and distribution; it causes surface buildings to tilt and collapse, damages infrastructure such as roads, bridges, and pipelines, and induces geological disasters such as landslides, mudslides, and ground fissures.

[0003] Mining area settlement prediction can provide early warning of potential settlement risks, promote the rational use of natural resources, promote ecological environmental protection, support geological disaster safety assessment, and provide an important basis for scientific decision-making. The prediction technologies currently used mainly include probability integral method, numerical simulation method, InSAR monitoring and prediction technology, typical curve method and profile function method, similar material simulation method, artificial neural network prediction method, etc.

[0004] The accuracy of the prediction results of the probability integral method and the numerical simulation method is controlled by the selection of parameters. Inappropriate parameters may lead to distortion of the prediction results. InSAR technology mainly detects surface deformation information and cannot directly solve underground problems such as underground rock displacement. The typical curve method and the profile function method need to establish a typical curve or function model based on a large amount of measured data. Therefore, they are highly targeted and not suitable for all geological and mining conditions. The similar material simulation method can usually only meet the requirements of qualitative or semi-quantitative analysis and is difficult to meet the needs of high-precision prediction. The artificial neural network algorithm is prone to fall into the local optimal solution during the training process, causing the prediction results to deviate from the actual situation. Summary of the invention

[0005] In view of the above problems, the present invention proposes a geological method and a geological device for intelligent prediction of coal mining subsidence.

[0006] The embodiment of the present invention provides a geological method for intelligent prediction of coal mining subsidence, and the geological method for intelligent prediction of coal mining subsidence includes:

[0007] Collect geological data that restricts surface subsidence in the mining area, including: geological information, hydrological information and mining activity information of the coal mining area;

[0008] Use InSAR monitoring technology to obtain high-precision time series data of surface deformation in the mining area;

[0009] Combining the geological data with the surface deformation time series data, a kernel-based algorithm is used to predict coal mining subsidence.

[0010] Optionally, the coal mining area geological information includes but is not limited to the stratum information and coal seam information of the mining area;

[0011] The stratum information includes the lithology, occurrence, thickness, rock physical and mechanical properties and solidity coefficient of the overlying strata in the mining area, and the thickness of the loose layer;

[0012] The coal seam information includes coal seam thickness and coal seam inclination;

[0013] The hydrological information includes the change of groundwater and the hydrogeological conditions of the loose layer;

[0014] The mining activity information includes working face layout information and roof and floor information;

[0015] The working face layout information includes the mining length, mining width, mining depth and mining height of the working face;

[0016] The roof and floor information includes the lithology and thickness of the old roof, immediate roof and immediate bottom of the working face.

[0017] Optionally, use InSAR monitoring technology to obtain high-precision time series data of surface deformation in the mining area, including:

[0018] Step D1: data preparation and preprocessing;

[0019] Step D2: differential interferometry processing;

[0020] Step D3: deformation analysis and result interpretation;

[0021] The surface deformation time series data include but are not limited to deformation amount and deformation direction, deformation rate and deformation time series information.

[0022] Optionally, the data preparation and preprocessing includes:

[0023] Step D11: Select appropriate radar satellite data according to the surface characteristics of the mining area to obtain radar satellite images;

[0024] Step D12: pre-processing the radar satellite image to obtain a single-view complex image;

[0025] Step D13: registering and resampling the single-view complex images;

[0026] The preprocessing of the radar satellite image includes: radiation correction, speckle noise filtering, geometric correction, image registration, phase correction, and image cropping and resampling.

[0027] Optionally, the differential interferometry processing includes:

[0028] Step D21: Select an appropriate time series InSAR method to generate an interferogram using the registered single-view complex images;

[0029] Step D22: performing interference processing and phase differentiation on the interference pattern to obtain the required deformation phase and obtain a differential interference pattern;

[0030] Step D23: filtering the differential interference pattern to improve the differential phase signal-to-noise ratio and reduce residual influence;

[0031] The interference processing and phase difference of the interference pattern is performed to separate the deformation phase from the interference phase by removing the influence of the flat ground effect and the terrain phase.

[0032] Optionally, the deformation analysis and result interpretation include:

[0033] Step D31: performing phase unwrapping on the filtered differential interferogram to extract surface deformation information;

[0034] Step D32: extracting the surface deformation information and performing geocoding;

[0035] Step D33: Perform deformation analysis and accuracy assessment after geocoding;

[0036] The phase unwrapping is performed by restoring the real phase value of the entangled interference phase to obtain continuous deformation phase information;

[0037] The geocoding is performed by converting the acquired deformation results from the radar coordinate system into a map projection coordinate system;

[0038] The deformation analysis and accuracy assessment are verified by combining the actual geological and surface conditions and GPS measured data, thereby obtaining high-precision surface deformation results and obtaining the surface deformation time series data.

[0039] The surface deformation time series data is a two-dimensional grid of time series, which is used to identify the change of the settlement value of each monitoring point in a specific area over time.

[0040] Optionally, a kernel-based algorithm is used to predict coal mining subsidence, including:

[0041] Step T1: data preprocessing and feature selection;

[0042] Step T2: kernel function selection and model training;

[0043] Step T3: Use the trained model to predict settlement for the new mining plan.

[0044] Optionally, the data preprocessing and feature selection include:

[0045] Step T11: Missing value and outlier processing:

[0046] If any satellite image in any revisit period is missing, time series interpolation is performed in the time dimension to ensure the continuity of monitoring data; if monitoring points in some areas are missing, the monitoring information of the missing points is interpolated in the spatial dimension according to the spatial correlation characteristics of surface subsidence;

[0047] Step T12: Feature extraction and selection:

[0048] Extracting displacement, displacement rate and acceleration data from the surface deformation time series data as main features; performing principal component analysis on various influencing factors in the geological data, and selecting the main influencing factors as external environment features of the prediction model;

[0049] Step T13: Data standardization processing:

[0050] In the case where the dimensions of different variables differ greatly, the surface deformation time series data and the geological data are subjected to Min-Max normalization or Z-score standardization processing;

[0051] Step T14: Feature analysis and data fusion:

[0052] According to the spatiotemporal characteristics of InSAR monitoring data, the external environment characteristics are matched to each grid point in the time dimension and space dimension according to static and dynamic characteristics to obtain fused data; the fused data is divided into multiple sub-areas, and the spatial adjacency relationship of each sub-area is obtained using the Delaunay triangulation network to construct the spatiotemporal feature vector of each monitoring point.

[0053] Optionally, the kernel function selection and model training include:

[0054] Step T21: Sample division and data set construction:

[0055] The wavelet transform method is used to decompose the spatiotemporal feature vector of each monitoring point into two subsequences of low frequency and high frequency. The time series segmentation method is used to construct the training data set and the test data set. GPS measurement data or leveling measurement data are selected as the verification data set.

[0056] The time series segmentation method is as follows: constructing a data set using a sliding window method, using the settlement data and influencing factors of the past T days to predict the settlement of the next 𝑡 days;

[0057] Step T22: Kernel method selection:

[0058] According to the complex nonlinear characteristics of the coal mining subsidence process, multiple kernel functions that capture nonlinear changes are selected, and according to the spatiotemporal characteristics of the InSAR monitoring data, multiple machine learning methods with time series regression advantages are selected to construct a combined model, and the model performance of different parameter combinations is evaluated by the grid search method and the particle swarm algorithm to obtain the optimal regression parameters;

[0059] The kernel functions include: polynomial kernel function, Gaussian radial basis kernel function and Matern kernel function;

[0060] The machine learning methods include: support vector regression, Gaussian process regression and kernelized neural network methods;

[0061] Step T23: Model training and hyperparameter optimization:

[0062] In the same mode, each combination model is trained according to the training data set and the test data set to obtain a prediction model after preliminary training, and a loss function is used to measure the error between the predicted value and the actual value, and different hyperparameter combinations are verified on the training set; based on the performance on the verification set, the Adam optimizer is selected to further obtain a hyperparameter combination with a smaller error;

[0063] Step T24: Model evaluation:

[0064] The cross-validation method is used to evaluate the performance and generalization ability of the model, and the mean square error, mean absolute error and determination coefficient indicators are used to evaluate the accuracy of the model. Based on the evaluation results, the optimal combined model is selected as the settlement prediction model for this mining area. On this basis, the training data set and test data set are further divided to optimize the model prediction performance, and the model is updated in a rolling manner to obtain the optimal prediction results.

[0065] Optionally, using the trained model to predict the settlement of a new mining scheme includes:

[0066] The trained model is used to predict the subsidence of new mining plans to obtain the predicted subsidence amount and subsidence range. The model is updated and improved online by continuously updating InSAR monitoring data to further improve the prediction accuracy.

[0067] The embodiment of the present invention provides a geological device for intelligent prediction of coal mining subsidence, and the geological device for intelligent prediction of coal mining subsidence comprises:

[0068] A data collection module is used to collect geological data that restricts surface subsidence in the mining area, wherein the geological data includes: geological information, hydrological information and mining activity information of the coal mining area;

[0069] Acquisition of settlement data module, used to obtain high-precision time series data of surface deformation in the mining area using InSAR monitoring technology;

[0070] A prediction module is used to combine the geological data and the surface deformation time series data to predict coal mining subsidence based on a kernel algorithm.

[0071] Optionally, the geological information of the coal mining area collected by the data collection module includes but is not limited to stratum information and coal seam information of the mining area;

[0072] The stratum information includes the lithology, occurrence, thickness, rock physical and mechanical properties and solidity coefficient of the overlying strata in the mining area, and the thickness of the loose layer;

[0073] The coal seam information includes coal seam thickness and coal seam inclination;

[0074] The hydrological information includes the change of groundwater and the hydrogeological conditions of the loose layer;

[0075] The mining activity information includes working face layout information and roof and floor information;

[0076] The working face layout information includes the mining length, mining width, mining depth and mining height of the working face;

[0077] The roof and floor information includes the lithology and thickness of the old roof, immediate roof and immediate bottom of the working face.

[0078] Optionally, the module for acquiring sedimentation data includes:

[0079] Processing submodule, used for data preparation and preprocessing;

[0080] Differential interferometry submodule, used for differential interferometry processing;

[0081] Analysis and interpretation submodule, used for deformation analysis and result interpretation;

[0082] The surface deformation time series data include but are not limited to deformation amount and deformation direction, deformation rate and deformation time series information.

[0083] Optionally, the processing submodule is specifically used to perform the following steps:

[0084] Step D11: Select appropriate radar satellite data according to the surface characteristics of the mining area to obtain radar satellite images;

[0085] Step D12: pre-processing the radar satellite image to obtain a single-view complex image;

[0086] Step D13: registering and resampling the single-view complex images;

[0087] The preprocessing of the radar satellite image includes: radiation correction, speckle noise filtering, geometric correction, image registration, phase correction, and image cropping and resampling.

[0088] Optionally, the differential interference submodule is specifically used to perform the following steps:

[0089] Step D21: Select an appropriate time series InSAR method to generate an interferogram using the registered single-view complex images;

[0090] Step D22: performing interference processing and phase differentiation on the interference pattern to obtain the required deformation phase and obtain a differential interference pattern;

[0091] Step D23: filtering the differential interference pattern to improve the differential phase signal-to-noise ratio and reduce residual influence;

[0092] The interference processing and phase difference of the interference pattern is performed to separate the deformation phase from the interference phase by removing the influence of the flat ground effect and the terrain phase.

[0093] Optionally, the analysis and interpretation submodule is specifically used to perform the following steps:

[0094] Step D31: performing phase unwrapping on the filtered differential interferogram to extract surface deformation information;

[0095] Step D32: extracting the surface deformation information and performing geocoding;

[0096] Step D33: Perform deformation analysis and accuracy assessment after geocoding;

[0097] The phase unwrapping is performed by restoring the real phase value of the entangled interference phase to obtain continuous deformation phase information;

[0098] The geocoding is performed by converting the acquired deformation results from the radar coordinate system into a map projection coordinate system;

[0099] The deformation analysis and accuracy assessment are verified by combining the actual geological and surface conditions and GPS measured data, thereby obtaining high-precision surface deformation results and obtaining the surface deformation time series data.

[0100] The surface deformation time series data is a two-dimensional grid of time series, which is used to identify the change of the settlement value of each monitoring point in a specific area over time.

[0101] Optionally, the prediction module includes:

[0102] Processing selection submodule, used for data preprocessing and feature selection;

[0103] Select the training submodule for kernel function selection and model training;

[0104] The subsidence prediction submodule is used to use the trained model to predict the subsidence of new mining plans.

[0105] Optionally, the processing selection submodule is specifically used to perform the following steps:

[0106] Step T11: Missing value and outlier processing:

[0107] If any satellite image in any revisit period is missing, time series interpolation is performed in the time dimension to ensure the continuity of monitoring data; if monitoring points in some areas are missing, the monitoring information of the missing points is interpolated in the spatial dimension according to the spatial correlation characteristics of surface subsidence;

[0108] Step T12: Feature extraction and selection:

[0109] Extracting displacement, displacement rate and acceleration data from the surface deformation time series data as main features; performing principal component analysis on various influencing factors in the geological data, and selecting the main influencing factors as external environment features of the prediction model;

[0110] Step T13: Data standardization processing:

[0111] In the case where the dimensions of different variables differ greatly, the surface deformation time series data and the geological data are subjected to Min-Max normalization or Z-score standardization processing;

[0112] Step T14: Feature analysis and data fusion:

[0113] According to the spatiotemporal characteristics of InSAR monitoring data, the external environment characteristics are matched to each grid point in the time dimension and space dimension according to static and dynamic characteristics to obtain fused data; the fused data is divided into multiple sub-areas, and the spatial adjacency relationship of each sub-area is obtained using the Delaunay triangulation network to construct the spatiotemporal feature vector of each monitoring point.

[0114] Optionally, the selection training submodule is specifically used to perform the following steps:

[0115] Step T21: Sample division and data set construction:

[0116] The wavelet transform method is used to decompose the spatiotemporal feature vector of each monitoring point into two subsequences of low frequency and high frequency. The time series segmentation method is used to construct the training data set and the test data set. GPS measurement data or leveling measurement data are selected as the verification data set.

[0117] The time series segmentation method is as follows: constructing a data set using a sliding window method, using the settlement data and influencing factors of the past T days to predict the settlement of the next 𝑡 days;

[0118] Step T22: Kernel method selection:

[0119] According to the complex nonlinear characteristics of the coal mining subsidence process, multiple kernel functions that capture nonlinear changes are selected, and according to the spatiotemporal characteristics of the InSAR monitoring data, multiple machine learning methods with time series regression advantages are selected to construct a combined model, and the model performance of different parameter combinations is evaluated by the grid search method and the particle swarm algorithm to obtain the optimal regression parameters;

[0120] The kernel functions include: polynomial kernel function, Gaussian radial basis kernel function and Matern kernel function;

[0121] The machine learning methods include: support vector regression, Gaussian process regression and kernelized neural network methods;

[0122] Step T23: Model training and hyperparameter optimization:

[0123] In the same mode, each combination model is trained according to the training data set and the test data set to obtain a prediction model after preliminary training, and a loss function is used to measure the error between the predicted value and the actual value, and different hyperparameter combinations are verified on the training set; based on the performance on the verification set, the Adam optimizer is selected to further obtain a hyperparameter combination with a smaller error;

[0124] Step T24: Model evaluation:

[0125] The cross-validation method is used to evaluate the performance and generalization ability of the model, and the mean square error, mean absolute error and determination coefficient indicators are used to evaluate the accuracy of the model. Based on the evaluation results, the optimal combined model is selected as the settlement prediction model for this mining area. On this basis, the training data set and test data set are further divided to optimize the model prediction performance, and the model is updated in a rolling manner to obtain the optimal prediction results.

[0126] Optionally, the settlement prediction submodule is specifically used for:

[0127] The trained model is used to predict the settlement of new mining plans and obtain the predicted settlement amount and range. The model is updated and improved online by continuously updating InSAR monitoring data to further improve the prediction accuracy.

[0128] The present invention discloses a geological method for intelligent prediction of coal mining subsidence, which comprises: collecting detailed geological data that restrict surface subsidence in the coal mining area, such as geological information, hydrological information and mining activities; using INSAR monitoring technology to obtain high-precision time series data of surface deformation in the mining area; and combining geological data and surface deformation data to predict coal mining subsidence based on a kernel-based algorithm.

[0129] The intelligent prediction geological method for coal mining subsidence provided by the present invention realizes the intelligent prediction and analysis of surface subsidence in mining areas, improves the accuracy and efficiency of coal mining subsidence prediction, provides a scientific basis for the control and management of coal mining subsidence, helps to reduce the impact of mining activities on the environment, and ensures the safety and sustainable development of coal mining. It has good application value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0130] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0131] Figure 1 A flow chart of a method for intelligent prediction of coal mining subsidence in an embodiment of the present invention;

[0132] Figure 2 The present invention is a block diagram of a geological device for intelligent prediction of coal mining subsidence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0133] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, are only part of the embodiments of the present invention, not all of the embodiments, and are not used to limit the present invention.

[0134] The intelligent geological prediction method for coal mining subsidence proposed in the present invention is based on Figure 1 The flowchart of the intelligent prediction geological method for coal mining subsidence is shown, and the intelligent prediction geological method for coal mining subsidence includes the following steps:

[0135] Step 101: Collect geological data that restricts surface subsidence in the mining area, including: geological information, hydrological information and mining activity information of the coal mining area.

[0136] The intelligent geological prediction method for coal mining subsidence provided by the present invention first needs to collect geological data that restrict the surface subsidence of the mining area, and the geological data includes: geological information, hydrological information and mining activity information of the coal mining area. The higher the accuracy of the geological data, the more accurate the subsequent prediction results of coal mining subsidence will be.

[0137] Preferably, the geological information of the coal mining area includes but is not limited to: the stratigraphic information and coal seam information of the mining area, wherein the stratigraphic information includes: the lithology, occurrence, thickness, rock physical and mechanical properties and strength coefficient of the overlying strata in the mining area, and the thickness of the loose layer on the surface; and the coal seam information includes: the thickness of the coal seam and the inclination of the coal seam.

[0138] Preferably, the hydrological information in the geological data includes: the change of groundwater and the hydrogeological conditions of the loose layer, both of which can be characterized by monitoring the height of the water level.

[0139] Preferably, the mining activity information in the geological data includes: working face layout information and roof and floor information, wherein the working face layout information includes: mining length, mining width, mining depth and mining height of the working face that has been constructed or is about to be constructed; and the roof and floor information includes: the lithology and thickness of the old roof, direct roof and direct bottom of the working face that has been constructed or is about to be constructed.

[0140] In addition, mining activity information can also include mining methods and roof management. Different mining methods have different effects on surface subsidence. For example, longwall mining is prone to cause large surface subsidence due to the large area of ​​the goaf; while improper roof management may cause the rock layer above the goaf to fall, aggravating surface subsidence.

[0141] Step 102: Use InSAR monitoring technology to obtain high-precision time series data of surface deformation in the mining area.

[0142] In addition to obtaining geological data, it is also necessary to use InSAR monitoring technology to obtain high-precision time series data of surface deformation in the mining area. The so-called InSAR monitoring technology, also known as synthetic aperture radar interferometry technology, is a technology that uses the interference image of SAR data to process phase information. Its purpose is to obtain fine surface height change information by measuring phase difference. The so-called SAR, that is, synthetic aperture radar, is an active microwave remote sensing technology that obtains surface information by transmitting and receiving microwave signals.

[0143] Among them, the surface deformation time series data usually refers to: a two-dimensional grid of the time series, which is used to identify the change of the settlement value of each monitoring point in a specific area over time.

[0144] A better method of using InSAR monitoring technology to obtain high-precision time series data of surface deformation in the mining area includes:

[0145] Step D1: data preparation and preprocessing;

[0146] Step D2: differential interferometry processing;

[0147] Step D3: deformation analysis and result interpretation;

[0148] The surface deformation time series data include but are not limited to: deformation amount and deformation direction, deformation rate and deformation time series information.

[0149] Preferably, the data preparation and preprocessing in the above step D1 may include the following steps:

[0150] Step D11: Select appropriate radar satellite data according to the surface characteristics of the mining area to obtain radar satellite images.

[0151] In fact, radar satellite data include but are not limited to: TerraSAR-X, RADARSAT, Sentinel-1. When selecting radar satellite data, it is necessary to consider the satellite's revisit period, band, resolution and other parameters to ensure that the selected data can meet the monitoring needs.

[0152] Since the period of coal mining settlement is relatively long, generally one to two years, satellites with a revisit period of several hours to several days can meet the needs. The consideration of band parameters can be selected according to specific circumstances. For example, for areas with sparse vegetation, the X-band with a shorter wavelength and sensitive phase change can be selected for high-precision deformation monitoring, while for areas with more vegetation coverage, the C-band with a longer wavelength and stronger penetration can be selected for surface deformation monitoring. Selecting appropriate radar satellite data can obtain radar satellite images of the mining area.

[0153] After step D11 is completed, step D12 is performed: pre-processing the radar satellite image to obtain a single-view complex image; and step D13 is performed: registering and resampling the single-view complex image.

[0154] Among them, the preprocessing of radar satellite images includes: radiation correction, speckle noise filtering, geometric correction, image registration, phase correction, and image cropping and resampling.

[0155] Among them, radiation correction is used to correct various instrument errors within the SAR system, and speckle noise filtering is performed at the same time to ensure that the SAR image reflects the characteristics of the ground objects more accurately; geometric correction is used to remove the geometric distortion caused by terrain undulations to ensure that the ground objects and image pixels match more accurately; image registration ensures that the pixels of SAR images acquired at different times are completely aligned; phase correction is used to remove the influence of atmospheric effects and orbit errors on the subsequent interferogram generation.

[0156] The purpose of step D13: registering and resampling the single-view complex images is to achieve spatial alignment of images from different periods.

[0157] Preferably, the differential interference processing in the above step D2 may include the following steps:

[0158] Step D21: Select an appropriate time series InSAR method to generate an interferogram using the registered single-look complex images.

[0159] The so-called time series InSAR method includes but is not limited to: PS-InSAR method, SBAS_InSAR method and DS-InSAR method. Generally, a suitable time series InSAR method can be selected according to the characteristics of the selected satellite image and the features of the ground objects in the area where the image is obtained.

[0160] Step D22: performing interference processing and phase differentiation on the interference pattern to obtain the required deformation phase and obtain a differential interference pattern.

[0161] The interference pattern is subjected to interference processing and phase differentiation, the goal of which is to separate the deformation phase from the interference phase by removing the influence of the flat ground effect and the terrain phase.

[0162] Step D23: Filter the differential interference pattern to improve the differential phase signal-to-noise ratio and reduce the residual effect.

[0163] The purpose of filtering the differential interferogram is to further remove the influence of the residual phase and reduce the difficulty of phase unwrapping.

[0164] Preferably, the deformation analysis and result interpretation in the above step D3 may include the following steps:

[0165] Step D31: performing phase unwrapping on the filtered differential interferogram to extract surface deformation information.

[0166] The purpose of phase unwrapping of the filtered differential interferogram is to obtain continuous deformation phase information by restoring the true phase value of the wrapped interference phase.

[0167] Step D32: extracting the surface deformation information and performing geocoding.

[0168] The purpose of geocoding is to transform the acquired deformation results from the radar coordinate system into the map projection coordinate system.

[0169] Step D33: Perform deformation analysis and accuracy assessment after geocoding.

[0170] Deformation analysis and accuracy assessment refers to obtaining high-precision surface deformation results by combining geological and surface conditions with GPS measured data for verification. This is to explain the deformation results and ensure the reliability of the monitoring results.

[0171] Step 103: Combine geological data and surface deformation time series data to predict coal mining subsidence based on a kernel algorithm.

[0172] After obtaining geological data through the aforementioned step 101 and surface deformation time series data through the aforementioned step 102, coal mining subsidence can be predicted by combining the two and then using a kernel-based algorithm.

[0173] A better method for predicting coal mining subsidence based on kernel algorithm includes the following steps:

[0174] Step T1: data preprocessing and feature selection;

[0175] Step T 2: kernel function selection and model training;

[0176] Step T 3: Use the trained model to predict the settlement of the new mining plan.

[0177] Preferably, the data preprocessing and feature selection in the above step T1 may include the following steps:

[0178] Step T11: Processing of missing values ​​and outliers: If any satellite image in any revisit period is missing, time series interpolation is performed in the time dimension to ensure the continuity of the monitoring data; if monitoring points in some areas are missing, the monitoring information of the missing points is interpolated in the spatial dimension based on the spatial correlation characteristics of surface subsidence.

[0179] Step T12 Feature extraction and selection: Extract the displacement, displacement rate and acceleration data in the surface deformation time series data as the main features; perform principal component analysis on the influencing factors in the geological data, and select the main influencing factors as the external environment characteristics of the prediction model.

[0180] Step T13: Data standardization: When the dimensions of different variables differ greatly, Min-Max normalization or Z-score standardization is performed on the surface deformation time series data and geological data.

[0181] Step T14 feature analysis and data fusion: According to the spatiotemporal characteristics of InSAR monitoring data, the external environment characteristics are matched to each grid point in the time dimension and space dimension according to the static and dynamic characteristics to obtain fused data; the fused data is divided into multiple sub-areas, and the Delaunay triangulation is used to obtain the spatial adjacency relationship of each sub-area, and the spatiotemporal feature vector of each monitoring point is constructed.

[0182] Preferably, the kernel function selection and model training in the above step T2 may include the following steps:

[0183] Step T21: Sample division and data set construction: Wavelet transform method is used to decompose the spatiotemporal feature vector of each monitoring point into two subsequences of low frequency and high frequency. Time series segmentation method is used to construct training data set and test data set, and GPS measurement data or leveling measurement data is selected as verification data set. The so-called time series segmentation method refers to: using the sliding window method to construct a data set, using the settlement data and influencing factors of the past T days to predict the settlement of the next 𝑡 days.

[0184] Step T22 Kernel method selection: According to the complex nonlinear characteristics of the coal mining subsidence process, multiple kernel functions that capture nonlinear changes are selected, and according to the spatiotemporal characteristics of the InSAR monitoring data, multiple machine learning methods with the advantage of time series regression are selected to build a combined model. The model performance of different parameter combinations is evaluated through the grid search method and the particle swarm algorithm to obtain the optimal regression parameters; the kernel functions include polynomial kernel function, Gaussian kernel function and Matern kernel function; the machine learning methods include support vector regression (SVR), Gaussian process regression (GPR) and kernelized neural network (KNN) method.

[0185] Step T23: Model training and hyperparameter optimization: In the same mode, each combination model is trained according to the training data set and the test data set to obtain the prediction model after preliminary training. The loss function is used to measure the error between the predicted value and the actual value, and the combination of different hyperparameters is verified on the training set. According to the performance on the verification set, the Adam optimizer is selected to further obtain the hyperparameter combination with smaller error.

[0186] Step T24 Model evaluation: Use the cross-validation method to evaluate the performance and generalization ability of the model, and use the mean square error (MSE), mean absolute error (MAE) and determination coefficient (R²) indicators to evaluate the accuracy of the model. According to the evaluation results, select the best combination model as the settlement prediction model for this mining area, and further divide the training data set and test data set on this basis to optimize the model prediction performance, and update it in a rolling manner to obtain the best prediction results.

[0187] . Preferably, the above step T3 using the trained model to predict the settlement of the new mining scheme may include: inputting characteristic data such as geological information, hydrological information and mining activity information of the coal mining area corresponding to the new mining scheme (that is, geological data of the coal mining area corresponding to the new mining scheme) into the trained model, predicting the settlement of the new mining scheme, and thus obtaining the predicted settlement amount and settlement range. By continuously updating the InSAR monitoring data, the model can be updated and improved online to further improve the prediction accuracy.

[0188] Based on the above-mentioned coal mining settlement intelligent prediction geological method, the embodiment of the present invention also provides a coal mining settlement intelligent prediction geological device, referring to Figure 2 The block diagram of the intelligent prediction device for coal mining subsidence shown in FIG. 1 includes:

[0189] The data collection module 210 is used to collect geological data that restricts the surface subsidence of the mining area, and the geological data includes: geological information, hydrological information and mining activity information of the coal mining area;

[0190] A settlement data acquisition module 220 is used to acquire high-precision time series data of surface deformation in the mining area using InSAR monitoring technology;

[0191] The prediction module 230 is used to combine the geological data and the surface deformation time series data to predict coal mining subsidence based on a kernel algorithm.

[0192] Optionally, the geological information of the coal mining area collected by the data collection module 210 includes but is not limited to the stratum information and coal seam information of the mining area;

[0193] The stratum information includes the lithology, occurrence, thickness, rock physical and mechanical properties and solidity coefficient of the overlying strata in the mining area, and the thickness of the loose layer;

[0194] The coal seam information includes coal seam thickness and coal seam inclination;

[0195] The hydrological information includes the change of groundwater and the hydrogeological conditions of the loose layer;

[0196] The mining activity information includes working face layout information and roof and floor information;

[0197] The working face layout information includes the mining length, mining width, mining depth and mining height of the working face;

[0198] The roof and floor information includes the lithology and thickness of the old roof, immediate roof and immediate bottom of the working face.

[0199] Optionally, the module 220 for acquiring sedimentation data includes:

[0200] Processing submodule, used for data preparation and preprocessing;

[0201] Differential interferometry submodule, used for differential interferometry processing;

[0202] Analysis and interpretation submodule, used for deformation analysis and result interpretation;

[0203] The surface deformation time series data include but are not limited to deformation amount and deformation direction, deformation rate and deformation time series information.

[0204] Optionally, the processing submodule is specifically used to perform the following steps:

[0205] Step D11: Select appropriate radar satellite data according to the surface characteristics of the mining area to obtain radar satellite images;

[0206] Step D12: pre-processing the radar satellite image to obtain a single-view complex image;

[0207] Step D13: registering and resampling the single-view complex images;

[0208] The preprocessing of the radar satellite image includes: radiation correction, speckle noise filtering, geometric correction, image registration, phase correction, and image cropping and resampling.

[0209] Optionally, the differential interference submodule is specifically used to perform the following steps:

[0210] Step D21: Select an appropriate time series InSAR method to generate an interferogram using the registered single-view complex images;

[0211] Step D22: performing interference processing and phase differentiation on the interference pattern to obtain the required deformation phase and obtain a differential interference pattern;

[0212] Step D23: filtering the differential interference pattern to improve the differential phase signal-to-noise ratio and reduce residual influence;

[0213] The interference processing and phase difference of the interference pattern is performed to separate the deformation phase from the interference phase by removing the influence of the flat ground effect and the terrain phase.

[0214] Optionally, the analysis and interpretation submodule is specifically used to perform the following steps:

[0215] Step D31: performing phase unwrapping on the filtered differential interferogram to extract surface deformation information;

[0216] Step D32: extracting the surface deformation information and performing geocoding;

[0217] Step D33: Perform deformation analysis and accuracy assessment after geocoding;

[0218] The phase unwrapping is performed by restoring the real phase value of the entangled interference phase to obtain continuous deformation phase information;

[0219] The geocoding is performed by converting the acquired deformation results from the radar coordinate system into a map projection coordinate system;

[0220] The deformation analysis and accuracy assessment are verified by combining the actual geological and surface conditions and GPS measured data, thereby obtaining high-precision surface deformation results and obtaining the surface deformation time series data.

[0221] The surface deformation time series data is a two-dimensional grid of time series, which is used to identify the change of the settlement value of each monitoring point in a specific area over time.

[0222] Optionally, the prediction module 230 includes:

[0223] Processing selection submodule, used for data preprocessing and feature selection;

[0224] Select the training submodule for kernel function selection and model training;

[0225] The subsidence prediction submodule is used to use the trained model to predict the subsidence of new mining plans.

[0226] Optionally, the processing selection submodule is specifically used for:

[0227] Step T11: Missing value and outlier processing:

[0228] If any satellite image in any revisit period is missing, time series interpolation is performed in the time dimension to ensure the continuity of monitoring data; if monitoring points in some areas are missing, the monitoring information of the missing points is interpolated in the spatial dimension according to the spatial correlation characteristics of surface subsidence;

[0229] Step T12: Feature extraction and selection:

[0230] Extracting displacement, displacement rate and acceleration data from the surface deformation time series data as main features; performing principal component analysis on various influencing factors in the geological data, and selecting the main influencing factors as external environment features of the prediction model;

[0231] Step T13: Data standardization processing:

[0232] In the case where the dimensions of different variables differ greatly, the surface deformation time series data and the geological data are subjected to Min-Max normalization or Z-score standardization processing;

[0233] Step T14: Feature analysis and data fusion:

[0234] According to the spatiotemporal characteristics of InSAR monitoring data, the external environment characteristics are matched to each grid point in the time dimension and space dimension according to static and dynamic characteristics to obtain fused data; the fused data is divided into multiple sub-areas, and the spatial adjacency relationship of each sub-area is obtained using the Delaunay triangulation network to construct the spatiotemporal feature vector of each monitoring point.

[0235] Optionally, the selection training submodule is specifically used for:

[0236] Step T21: Sample division and data set construction:

[0237] The wavelet transform method is used to decompose the spatiotemporal feature vector of each monitoring point into two subsequences of low frequency and high frequency. The time series segmentation method is used to construct the training data set and the test data set. GPS measurement data or leveling measurement data are selected as the verification data set.

[0238] Among them, the time series segmentation method is: use the sliding window method to construct a data set, use the settlement data and influencing factors of the past T days to predict the settlement of the next 𝑡 days.

[0239] Step T22: Kernel method selection:

[0240] According to the complex nonlinear characteristics of the coal mining subsidence process, multiple kernel functions that capture nonlinear changes are selected, and according to the spatiotemporal characteristics of the InSAR monitoring data, multiple machine learning methods with time series regression advantages are selected to construct a combined model, and the model performance of different parameter combinations is evaluated by the grid search method and the particle swarm algorithm to obtain the optimal regression parameters;

[0241] The kernel functions include: polynomial kernel function, Gaussian radial basis kernel function and Matern kernel function;

[0242] The machine learning methods include: support vector regression, Gaussian process regression and kernelized neural network methods;

[0243] Step T23: Model training and hyperparameter optimization:

[0244] In the same mode, each combination model is trained according to the training data set and the test data set to obtain a prediction model after preliminary training, and a loss function is used to measure the error between the predicted value and the actual value, and different hyperparameter combinations are verified on the training set; based on the performance on the verification set, the Adam optimizer is selected to further obtain a hyperparameter combination with a smaller error;

[0245] Step T24: Model evaluation:

[0246] The cross-validation method is used to evaluate the performance and generalization ability of the model, and the mean square error, mean absolute error and determination coefficient indicators are used to evaluate the accuracy of the model. According to the evaluation results, the optimal combination model is selected as the settlement prediction model of this mining area. On this basis, the training data set and the test data set are further divided to optimize the model prediction performance, and the rolling update is performed to obtain the optimal prediction results.

[0247] Optionally, the settlement prediction submodule is specifically used for:

[0248] Use the trained model to predict the settlement of the new mining plan to obtain the predicted settlement amount and settlement range; and update and improve the model online by continuously updating the InSAR monitoring data to further improve the prediction accuracy. In summary, the present invention provides a geological method for intelligent prediction of coal mining settlement, which includes: collecting detailed geological information that restricts the surface settlement of the mining area, such as geological information, hydrological information and mining activities; using INSAR monitoring technology to obtain high-precision surface deformation time series data of the mining area; combining geological data and surface deformation data, and predicting coal mining settlement based on a kernel-based algorithm.

[0249] The intelligent prediction geological method for coal mining subsidence proposed in the present invention realizes the intelligent prediction and analysis of surface subsidence in mining areas, improves the accuracy and efficiency of coal mining subsidence prediction, provides a scientific basis for the control and management of coal mining subsidence, helps to reduce the impact of mining activities on the environment, and ensures the safety and sustainable development of coal mining. It has good application value and promotion prospects.

[0250] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0251] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0252] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A geological method for intelligent prediction of coal mining subsidence, characterized in that: The coal mining subsidence intelligent prediction geological method comprises: Collect geological data that restricts surface subsidence in the mining area, including: geological information, hydrological information and mining activity information of the coal mining area; Use InSAR monitoring technology to obtain high-precision time series data of surface deformation in the mining area; Combining the geological data and the surface deformation time series data, a kernel-based algorithm is used to predict coal mining subsidence; Among them, the kernel-based algorithm predicts coal mining settlement, including: step T1: data preprocessing and feature selection; step T2: kernel function selection and model training; step T3: using the trained model to predict settlement of new mining plans; Among them, the data preprocessing and feature selection include: Step T11: Missing value and outlier processing: If any revisit period satellite image is missing, time series interpolation is performed in the time dimension to ensure the continuity of the monitoring data; If there are missing monitoring points in some areas, the monitoring information of the missing points is interpolated in the spatial dimension according to the spatial correlation characteristics of surface subsidence; Step T12: Feature extraction and selection: Extract the displacement, displacement rate and acceleration data in the surface deformation time series data as the main features; Perform principal component analysis on the influencing factors in the geological data, and select the main influencing factors as the external environment of the prediction model. Features; Step T13: Data standardization processing: When the dimensions of different variables are quite different, the surface deformation time series data and the geological data are subjected to Min-Max normalization or Z-score standardization processing; Step T14: Feature analysis and data fusion: According to the spatiotemporal characteristics of InSAR monitoring data, the external environmental features are matched to each grid point in the time dimension and the space dimension according to the static and dynamic characteristics to obtain fused data; the fused data is divided into multiple sub-areas, and the spatial adjacency relationship of each sub-area is obtained by using the Delaunay triangulation network to construct the spatiotemporal feature vector of each monitoring point; The kernel function selection includes: Step T21: Sample division and data set construction: The wavelet transform method is used to decompose the spatiotemporal feature vector of each monitoring point into two subsequences of low frequency and high frequency. The time series segmentation method is used to construct the training data set and the test data set, and the GPS measurement data or leveling measurement data is selected as the verification data set; wherein, the time series segmentation method is: using the sliding window method to construct the data set, using the settlement data and influencing factors of the past T days to predict the settlement of the next 𝑡 days; Step T22: Kernel method selection: According to the complex nonlinear characteristics of the coal mining subsidence process, multiple kernel functions that capture nonlinear changes are selected, and according to the spatiotemporal characteristics of the InSAR monitoring data, multiple machine learning methods with time series regression advantages are selected to construct a combined model, and the model performance of different parameter combinations is evaluated by the grid search method and the particle swarm algorithm to obtain the optimal regression parameters; Among them, the kernel functions include: polynomial kernel function, Gaussian radial basis kernel function and Matern kernel function; the machine learning methods include: support vector regression, Gaussian process regression and kernelized neural network method.

2. The method for intelligent prediction of coal mining subsidence according to claim 1 is characterized in that: The geological information of the coal mining area includes but is not limited to the stratum information and coal seam information of the mining area; The stratum information includes the lithology, occurrence, thickness, rock physical and mechanical properties and solidity coefficient of the overlying strata in the mining area, and the thickness of the loose layer; The coal seam information includes coal seam thickness and coal seam inclination; The hydrological information includes the change of groundwater and the hydrogeological conditions of the loose layer; The mining activity information includes working face layout information and roof and floor information; The working face layout information includes the mining length, mining width, mining depth and mining height of the working face; The roof and floor information includes the lithology and thickness of the old roof, immediate roof and immediate bottom of the working face.

3. The method for intelligent prediction of coal mining subsidence according to claim 1 is characterized in that: Use InSAR monitoring technology to obtain high-precision time series data of surface deformation in the mining area, including: Step D1: data preparation and preprocessing; Step D2: differential interferometry processing; Step D3: deformation analysis and result interpretation; The surface deformation time series data include but are not limited to deformation amount and deformation direction, deformation rate and deformation time series information.

4. The method for intelligent prediction of coal mining subsidence according to claim 3 is characterized in that: The data preparation and preprocessing include: Step D11: Select appropriate radar satellite data according to the surface characteristics of the mining area to obtain radar satellite images; Step D12: pre-processing the radar satellite image to obtain a single-view complex image; Step D13: registering and resampling the single-view complex images; The preprocessing of the radar satellite image includes: radiation correction, speckle noise filtering, geometric correction, image registration, phase correction, and image cropping and resampling.

5. The method for intelligent prediction of coal mining subsidence according to claim 4 is characterized in that: The differential interferometry processing includes: Step D21: Select an appropriate time series InSAR method to generate an interferogram using the registered single-view complex images; Step D22: performing interference processing and phase differentiation on the interference pattern to obtain the required deformation phase and obtain a differential interference pattern; Step D23: filtering the differential interference pattern to improve the differential phase signal-to-noise ratio and reduce residual influence; The interference processing and phase difference of the interference pattern is performed to separate the deformation phase from the interference phase by removing the influence of the flat ground effect and the terrain phase.

6. The intelligent geological prediction method for coal mining subsidence according to claim 5 is characterized in that: The deformation analysis and result interpretation include: Step D31: performing phase unwrapping on the filtered differential interferogram to extract surface deformation information; Step D32: extracting the surface deformation information and performing geocoding; Step D33: Perform deformation analysis and accuracy assessment after geocoding; The phase unwrapping is performed by restoring the real phase value of the entangled interference phase to obtain continuous deformation phase information; The geocoding is performed by converting the acquired deformation results from the radar coordinate system into a map projection coordinate system; The deformation analysis and accuracy assessment are verified by combining the actual geological and surface conditions and GPS measured data, thereby obtaining high-precision surface deformation results and obtaining the surface deformation time series data; The surface deformation time series data is a two-dimensional grid of time series, which is used to identify the change of the settlement value of each monitoring point in a specific area over time.

7. The intelligent geological prediction method for coal mining subsidence according to claim 1 is characterized in that: The model training comprises: Step T23: Model training and hyperparameter optimization: In the same mode, each combination model is trained according to the training data set and the test data set to obtain a prediction model after preliminary training, and a loss function is used to measure the error between the predicted value and the actual value, and different hyperparameter combinations are verified on the training set; based on the performance on the verification set, the Adam optimizer is selected to further obtain a hyperparameter combination with a smaller error; Step T24: Model evaluation: The cross-validation method is used to evaluate the performance and generalization ability of the model, and the mean square error, mean absolute error and determination coefficient indicators are used to evaluate the accuracy of the model. Based on the evaluation results, the optimal combined model is selected as the settlement prediction model for this mining area. On this basis, the training data set and test data set are further divided to optimize the model prediction performance, and the model is updated in a rolling manner to obtain the optimal prediction results.

8. The method for intelligent prediction of coal mining subsidence according to claim 1, characterized in that: The use of the trained model to predict the settlement of a new mining scheme includes: The trained model is used to predict the subsidence of new mining plans to obtain the predicted subsidence amount and subsidence range. The model is updated and improved online by continuously updating InSAR monitoring data to further improve the prediction accuracy.

9. An intelligent geological prediction device for coal mining settlement, characterized in that: The coal mining subsidence intelligent prediction geological device comprises: A data collection module is used to collect geological data that restricts surface subsidence in the mining area, wherein the geological data includes: geological information, hydrological information and mining activity information of the coal mining area; Acquisition of settlement data module, used to obtain high-precision time series data of surface deformation in the mining area using InSAR monitoring technology; A prediction module, used to predict coal mining subsidence based on a kernel-based algorithm by combining the geological data and the surface deformation time series data; The processing selection submodule is specifically used to perform the following steps: step T1: data preprocessing and feature selection; step T2: kernel function selection and model training; step T3: using the trained model to predict the settlement of the new mining plan; Among them, the data preprocessing and feature selection include: Step T11: Missing value and outlier processing: If any revisit period satellite image is missing, time series interpolation is performed in the time dimension to ensure the continuity of the monitoring data; If there are missing monitoring points in some areas, the monitoring information of the missing points is interpolated in the spatial dimension according to the spatial correlation characteristics of surface subsidence; Step T12: Feature extraction and selection: Extract the displacement, displacement rate and acceleration data in the surface deformation time series data as the main features; Perform principal component analysis on the influencing factors in the geological data, and select the main influencing factors as the external environment of the prediction model. Features; Step T13: Data standardization processing: When the dimensions of different variables are quite different, the surface deformation time series data and the geological data are subjected to Min-Max normalization or Z-score standardization processing; Step T14: Feature analysis and data fusion: According to the spatiotemporal characteristics of InSAR monitoring data, the external environmental features are matched to each grid point in the time dimension and the space dimension according to the static and dynamic characteristics to obtain fused data; the fused data is divided into multiple sub-areas, and the spatial adjacency relationship of each sub-area is obtained by using the Delaunay triangulation network to construct the spatiotemporal feature vector of each monitoring point; Among them, the selected training submodule is specifically used to perform the following steps: Step T21: Sample division and data set construction: The wavelet transform method is used to decompose the spatiotemporal feature vector of each monitoring point into two subsequences of low frequency and high frequency, and the time series segmentation method is used to construct a training data set and a test data set, and GPS measurement data or leveling measurement data is selected as a verification data set; wherein, the time series segmentation method is: using a sliding window method to construct a data set, using the settlement data and influencing factors of the past T days to predict the settlement of the next 𝑡 days; Step T22: Kernel method selection: According to the complex nonlinear characteristics of the coal mining settlement process, multiple kernel functions that capture nonlinear changes are selected, and according to the spatiotemporal characteristics of the InSAR monitoring data, multiple machine learning methods with time series regression advantages are selected to construct a combined model, and the model performance of different parameter combinations is evaluated by the grid search method and the particle swarm algorithm to obtain the optimal regression parameters; wherein, the kernel functions include: polynomial kernel function, Gaussian radial basis kernel function and Matern kernel function; the machine learning methods include: support vector regression, Gaussian process regression and kernelized neural network method.

Images