Intelligent analysis method and system for geological exploration data of coal mining subsidence area

Abstract

The invention relates to the field of geological exploration, in particular to an intelligent analysis method and system for geological exploration data of a coal mining subsidence area. The invention discloses an intelligent analysis system for geological exploration data of a coal mining subsidence area. The intelligent analysis system comprises a data acquisition module, an underground three-dimensional model construction module, a foundation stability evaluation module, a residual deformation prediction module and a suitability judgment module. According to the method, multi-source data such as drilling data, geophysical prospecting data, remote sensing topographic data, monitoring actual measurement data and geotechnical test data are comprehensively analyzed, a unified underground three-dimensional model is constructed, and the limitation of data isolation of a traditional exploration method is broken through; according to the method, through drilling and geophysical prospecting bidirectional complementary correction and combination of a settlement monitoring sequence and static attributes, a more accurate and comprehensive geological exploration result is provided for a coal mining subsidence area, and the accuracy of foundation stability evaluation and deformation prediction is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of geological exploration, specifically to an intelligent analysis method and system for geological exploration data in coal mining subsidence areas. Background Technology

[0002] As coal mining progresses, the geological exploration of coal mining subsidence areas is attracting increasing attention. Coal mining activities, conducted deep underground, inevitably cause surface subsidence and deformation. These changes not only affect the stability of surface buildings but may also impact groundwater resources, the ecological environment, and the safety of surrounding infrastructure. Therefore, how to conduct effective geological exploration in coal mining subsidence areas and assess the stability of the regional foundation and predict deformation has become an important issue in mineral resource extraction and environmental protection.

[0003] Traditional methods for exploring coal mining subsidence areas mainly rely on technologies such as drilling, geophysical exploration, and remote sensing. However, these methods are often independent and lack systematic integration, resulting in insufficient data accuracy and comprehensive information. In the process of collecting multi-source data, different types of data have different scales, accuracies, and spatial distributions. How to effectively integrate these data and construct accurate underground three-dimensional models based on them is an important research challenge.

[0004] Furthermore, existing foundation stability analysis technologies typically rely on manual experience or single data analysis methods, lacking scientific quantitative methods to predict foundation stability, residual deformation, and the impact of loads. Based on this, how to utilize intelligent analysis methods, combined with multi-source data such as ground settlement monitoring data, three-dimensional underground models, geotechnical test data, and historical load data, to comprehensively assess the suitability for construction in coal mining subsidence areas has become an important direction for improving the safety of coal mining area construction.

[0005] This invention proposes an intelligent analysis method and system for geological exploration data in coal mining subsidence areas. By integrating multi-source data and utilizing physical models and machine learning algorithms, it solves the problems of data fusion, accuracy improvement, and comprehensive evaluation in traditional technologies. It can provide more accurate foundation stability assessment and deformation prediction, thereby providing a scientific basis for the construction and management of coal mining subsidence areas. Summary of the Invention

[0006] This invention integrates and analyzes multi-source data, including drilling data, geophysical data, remote sensing topographic data, monitoring and measurement data, and geotechnical test data, to construct a unified three-dimensional underground model, overcoming the limitations of isolated data in traditional exploration methods. This method provides more accurate and comprehensive geological exploration results for coal mining subsidence areas through bidirectional complementary correction of drilling and geophysical data, and the combination of settlement monitoring sequences and static attributes, significantly improving the accuracy of foundation stability assessment and deformation prediction.

[0007] A method for intelligent analysis of geological exploration data in coal mining subsidence areas includes:

[0008] The drilling data, geophysical data, remote sensing topographic data, monitoring and measured data, geotechnical test data, and historical data of the target coal mining subsidence area are acquired. The monitoring and measured data includes GNSS displacement time series, leveling displacement time series, and InSAR displacement time series, and the above displacement time series are defined as settlement monitoring series. The remote sensing topographic data provides topographic parameters, and the historical data provides mining depth, mining thickness, and building load parameters.

[0009] Two-way complementary correction is performed based on drilling and geophysical data: the initial geophysical inversion body is anchored by the borehole strata and amplitude, point-to-surface residual iterative correction is established, and attribute completion is performed on voxels with low core recovery rate and broken blind sections; convergence is performed by confidence weight, and a unified underground 3D model is output, including goaf boundary, coal pillar geometry, overburden failure zone, porosity distribution and residual confidence field, and borehole repair suggestions are generated;

[0010] By using the coal pillar size, porosity and overburden structure information in the unified underground three-dimensional model, and combining the strength, deformation and permeability parameters in the geotechnical test data, the safety factor of the coal pillar, the overburden failure height and the filling degree are calculated to obtain the foundation stability level.

[0011] Using the settlement monitoring sequence as the time-series input and the static attributes extracted from the unified underground 3D model, geotechnical test, remote sensing topography and historical data as auxiliary input, the feature influence weights are calculated by physical sensitivity, SHAP contribution and data confidence, and the subsidence, tilt and horizontal deformation fields of future periods are output through the residual deformation prediction model.

[0012] The influence depth of building load is calculated based on historical data or design parameters, and coupled with the overburden failure height and minimum mining depth given by the unified underground three-dimensional model to obtain the load risk level. The foundation stability level, residual deformation results and load risk are input into the rule engine to automatically generate suitable construction, restricted construction, and prohibited construction zones, as well as supporting treatment and design suggestions, to form a conclusion on the suitability of construction in the coal mining subsidence area.

[0013] Preferably, the specific operation of performing bidirectional complementary correction based on drilling and geophysical data is as follows:

[0014] Step S1: Using geophysical data as input, perform denoising, static correction, amplitude normalization, and gridding on a unified coordinate system and time base to generate a preliminary geophysical inversion body;

[0015] Step S2: Using borehole strata, elevation, and borehole properties as anchor points, establish a time-depth comparison table segment by segment along the borehole depth to correct the depth positioning of the inverted body; within the comparison window, linearly scale and recalibrate the voxel amplitude of the window according to the ratio of borehole property values ​​to geophysical values ​​to obtain the calibrated geophysical inversion body;

[0016] Step S3: Calculate the difference between the calibrated inversion body and the measured attributes at each borehole location to form a residual set; obtain the strike and dip of the coal seam based on the existing borehole elevation points and interpretation results, and construct an anisotropic search ellipsoid with strike as the major axis and dip as the minor axis; use anisotropic inverse distance weighted diffusion residuals within the ellipsoid to obtain the residual field, and update the geophysical inversion body in combination with preset relaxation coefficients;

[0017] Step S4: Scoring of drilling data, calculating scores for core recovery rate, well inclination, and logging completeness, and standardizing them to obtain a score value for each borehole; calculating scores for coverage quality, signal-to-noise ratio, and inversion fitting error for geophysical voxels, and standardizing them to obtain a score value for each voxel; standardizing all scores to the same interval through a unified interval mapping and segmented threshold table.

[0018] Select several calibration boreholes and corresponding geophysical voxels within the project area to form a calibration subset. Iterate through the preset coefficient combinations on the calibration subset and execute steps S2 to S3 to calculate the weighted error. Select the borehole weight and voxel weight corresponding to the minimum error as a fixed combination. When interpolating the residuals, the product of the distance attenuation weight and the total borehole weight is used as the comprehensive weight. The voxel update range is determined by the total voxel weight according to the graded relaxation coefficient table.

[0019] Step S5: When there are borehole segments in the drilling data with a core recovery rate lower than the preset core recovery rate threshold and marked as broken, select high-weight reliable boreholes in the same stratum according to the project's preset stratum coding rules. First, use the linear interpolation of the two nearest boreholes as the initial value, and then perform distance weighting and smoothing on the stratum plane according to the search radius. Merge the results into the final complete attributes according to the preset ratio and assign high, medium, and low confidence levels. When the borehole trajectory deviates from the plane or depth of the geophysical anomaly center by more than the threshold, correct the coordinates and sampling depth by the same amount according to the deviation direction and record it.

[0020] Step S6: Repeat steps S2 to S5. After each iteration, calculate the weighted root mean square error index and compare it with a preset threshold. The weighted root mean square error index is the square of the difference between the geophysical inversion body and the measured attributes of each borehole. It is calculated by weighting according to the weight of the borehole, and the average value is obtained and the square root is taken. If the weighted root mean square error is less than the preset threshold, or the decrease ratio of the error between two adjacent rounds is less than the preset threshold, or the number of iterations reaches the upper limit, the iteration is terminated. After termination, a unified underground 3D model is output, and a residual confidence field is output. Based on the high confidence region in the residual confidence field, the geometric center of the connected region is extracted, the plane coordinates of the fill hole are determined, a safety margin is reserved, and a list of fill hole suggestions is generated.

[0021] Preferably, by using the coal pillar size, porosity, and overburden structure information in a unified underground 3D model, combined with the strength, deformation, and permeability parameters from geotechnical test data, the safety factor of the coal pillar, the overburden failure height, and the filling degree are calculated to obtain the foundation stability level. The specific operation is as follows:

[0022] Extract the planar boundary, column width, column height, and overburden thickness at the center point of the coal pillar from the 3D model; extract the porosity distribution of the goaf and divide it into blocks according to connectivity; extract uniaxial compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, and natural unit weight of rock from the geotechnical test data, and classify the roof lithology into hard, medium, and relatively soft roofs according to a preset threshold table;

[0023] The load and average stress of the coal pillar are calculated using the load area method. Specifically, the load area of ​​the coal pillar is delineated by drawing a perpendicular line between the plane boundary of the coal pillar and the boundary of the adjacent roadway or goaf. The product of the overburden thickness and the natural unit weight of the rock is taken as the vertical overburden stress, which is then multiplied by the load area to obtain the load on the coal pillar. The average stress of the coal pillar is obtained from the plane area of ​​the coal pillar. The pillar strength reduction method is used to correct the uniaxial compressive strength according to factors such as the pillar width-to-height ratio, calculate the ultimate bearing capacity of the coal pillar, and calculate the ratio of the actual load to the ultimate bearing capacity of the coal pillar to obtain the safety factor of the coal pillar.

[0024] The overburden failure height is calculated based on the roof lithology and mining thickness; the goaf filling degree is calculated by weighted average of porosity, and when multiple coal seams exist, the calculation is performed layer by layer and weighted composite.

[0025] Stability is determined according to preset classification rules: when the safety factor of the coal pillar, the overburden failure height and the filling degree all meet the stability conditions, it is determined to be stable; otherwise, it is basically stable or unstable, and the corresponding stability level is output.

[0026] Preferably, the feature influence weights are calculated based on physical sensitivity, SHAP contribution, and data confidence, as follows:

[0027] Under a unified coordinate and time datum, a feature set is constructed for each monitoring point and fixed time window, including settlement monitoring sequence, static attributes of three-dimensional model, geotechnical test data, remote sensing topographic data and historical data;

[0028] Using the improved probability integral method as the baseline model, the cumulative settlement index of the baseline is calculated; positive and negative perturbations are applied to each feature, and the change in the cumulative settlement index of the baseline before and after the perturbation is calculated. The absolute values ​​of the positive and negative perturbation changes are compared, and the larger absolute value is taken as the original value of the physical sensitivity of the feature. The sensitivity score is obtained by standardizing to [0,1].

[0029] Train a gradient boosting tree surrogate model, calculate the SHAP value for each feature based on the model, and obtain the SHAP contribution score after standardization;

[0030] Coverage, noise, and stability scores are calculated for time-series elements, while voxel total weights and residual confidence scores are calculated for static elements. After standardization, time-series and static confidence scores are synthesized separately.

[0031] Finally, based on a fixed coefficient combination, the physical sensitivity score, SHAP contribution score, and data confidence score are combined to form the influence weight of each feature.

[0032] Preferably, the residual deformation prediction model is based on an LSTM model and includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer. The input layer receives the settlement monitoring sequence and static attributes and normalizes them. The LSTM layer processes the time series data and captures the long-term dependencies of the time series information. The attention mechanism layer calculates a weighted sum based on the influence weight of each feature to enhance the focus on key features for residual deformation prediction. The fully connected layer performs nonlinear mapping on the features processed by the LSTM layer and the attention mechanism layer and performs feature dimensionality reduction. The output layer generates predicted values ​​for residual settlement, tilt, and horizontal deformation in future time periods.

[0033] Preferably, the influence depth is calculated based on the building load in historical data or design parameters, and coupled with the overburden failure height and minimum mining depth given by the unified underground three-dimensional model to obtain the load risk level. The specific operation is as follows:

[0034] Based on the building load type, design height, and foundation area, the additional stress on the underground structure is calculated using the foundation stress propagation theory, and the load influence depth is calculated based on the ratio of additional stress to self-weight stress.

[0035] Based on the overburden failure height and the minimum mining depth, determine the regional construction conditions: if the minimum mining depth is greater than the load influence depth plus the overburden failure height, it is a suitable construction area; if it is close to it, it is a restricted construction area, and reinforcement is recommended; if it is less than it, it is a prohibited construction area, and construction is prohibited or remediation is required.

[0036] Based on the coupling results of load influence depth, overburden failure height and minimum mining depth, combined with foundation stability and residual deformation assessment, the load risk level is output, which is divided into low risk, medium risk and high risk.

[0037] An intelligent analysis system for geological exploration data in coal mining subsidence areas includes:

[0038] The data acquisition module is used to acquire drilling data, geophysical data, remote sensing topographic data, monitoring and measurement data, geotechnical test data and historical data of the target coal mining subsidence area;

[0039] The underground 3D model building module is used to perform two-way complementary correction between drilling and geophysical data, output a unified underground 3D model, and generate a list of borehole repair suggestions.

[0040] The foundation stability assessment module is used to calculate indicators such as the coal pillar safety factor, overburden failure height and filling degree based on the coal pillar size, porosity and overburden structure information in the unified underground three-dimensional model, combined with the strength, deformation and permeability parameters in the geotechnical test data, to obtain the foundation stability level.

[0041] The residual deformation prediction module is used to output the subsidence, tilt, and horizontal deformation fields for future periods through the residual deformation prediction model.

[0042] The suitability assessment module is used to obtain the load risk level, generate suitable, restricted, and prohibited construction zones, and provide supporting governance and design suggestions to form a conclusion on the suitability of construction in coal mining subsidence areas.

[0043] The present invention has the following advantages:

[0044] 1. This invention integrates and analyzes multi-source data, including drilling data, geophysical data, remote sensing topographic data, monitoring and measured data, and geotechnical test data, to construct a unified three-dimensional underground model, overcoming the limitations of isolated data in traditional exploration methods. This method provides more accurate and comprehensive geological exploration results for coal mining subsidence areas through bidirectional complementary correction of drilling and geophysical data, and the combination of settlement monitoring sequences and static attributes, significantly improving the accuracy of foundation stability assessment and deformation prediction.

[0045] 2. This invention calculates the influence weight of features by physical sensitivity, SHAP contribution, and data confidence, and combines it with a residual deformation prediction model to predict future ground settlement, tilt, and horizontal deformation, fully considering the comprehensive impact of various factors on deformation. This intelligent analysis method can dynamically assess load, overburden failure, and foundation stability, providing a scientific basis for subsequent construction suitability judgment and load risk assessment, effectively reducing human bias and uncertainty in traditional exploration methods, and improving the reliability of prediction results. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the intelligent analysis system for geological exploration data in coal mining subsidence areas used in an embodiment of the present invention. Detailed Implementation

[0047] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0048] Example 1: An intelligent analysis method for geological exploration data in coal mining subsidence areas, comprising:

[0049] The drilling data, geophysical data, remote sensing topographic data, monitoring and measured data, geotechnical test data, and historical data of the target coal mining subsidence area are acquired. The monitoring and measured data includes GNSS displacement time series, leveling displacement time series, and InSAR displacement time series, and the above displacement time series are defined as settlement monitoring series. The remote sensing topographic data provides topographic parameters, and the historical data provides mining depth, mining thickness, and building load parameters.

[0050] Based on drilling and geophysical data, a two-way complementary correction between drilling and geophysical data is performed. Specifically, this includes: calibrating the initial geophysical inversion volume using borehole strata and amplitude as anchor points, and performing iterative correction by establishing point-surface residuals; simultaneously, attribute completion is performed on geophysical voxels with low core recovery rates and fractured blind sections; after convergence by confidence weight, a unified underground 3D model is output, including goaf boundaries, coal pillar geometry, overburden failure zones, continuous porosity distribution, and residual confidence fields, and a list of borehole repair suggestions is generated.

[0051] By using the coal pillar size, porosity and overburden structure information in the unified underground three-dimensional model, and combining the strength, deformation and permeability parameters in the geotechnical test data, the coal pillar safety factor, overburden failure height and filling degree and other indicators are calculated to obtain the foundation stability level, which provides constraints for subsequent residual deformation prediction and load judgment.

[0052] Using the settlement monitoring sequence as the time-series input and the static attributes extracted from the unified underground 3D model, geotechnical test, remote sensing topography and historical data as auxiliary input, the feature influence weights are calculated by physical sensitivity, SHAP contribution and data confidence, and the subsidence, tilt and horizontal deformation fields of future periods are output through the residual deformation prediction model.

[0053] The influence depth of building load is calculated based on historical data or design parameters, and coupled with the overburden failure height and minimum mining depth given by the unified underground three-dimensional model to obtain the load risk level. Finally, the foundation stability level, residual deformation results and load risk are input into the rule engine to automatically generate suitable construction, restricted construction, and prohibited construction zones, as well as supporting treatment and design suggestions, thereby forming a conclusion on the suitability of construction in coal mining subsidence areas.

[0054] The rules engine generates suitable, restricted, and prohibited partitioning results based on the following logic:

[0055] Suitable construction area: The foundation stability level is stable or basically stable, the residual deformation value is within the allowable range (such as settlement less than 50 mm), and the load risk level is low or medium risk; this area is suitable for conventional construction, and conventional anti-deformation measures should be considered in the design stage.

[0056] Restricted construction area: The foundation stability level is basically stable, the residual deformation value is moderate (if the settlement is between 50 and 100 mm), and the load risk level is medium risk; the area is suitable for construction, but additional reinforcement measures are required, such as using flexible foundations or deepening the foundation depth, to reduce the impact of foundation deformation on the building.

[0057] Restricted construction zone: The foundation stability level is unstable, the residual deformation value is large (such as settlement greater than 100 mm), and the load risk level is high risk; this area is not suitable for construction and construction prohibition measures should be taken, or large-scale foundation reinforcement, renovation or relocation of existing buildings should be carried out.

[0058] Based on the zoning, the rule engine automatically generates corresponding governance and design recommendations according to the following principles: For suitable construction areas, conventional foundation design is recommended, with the addition of necessary monitoring points to track future foundation deformation; for restricted construction areas, foundation reinforcement is recommended, such as grouting, deep foundation design, and flexible foundations, with adjustments to foundation type and depth based on remaining deformation prediction results; for prohibited construction areas, prohibition measures are recommended, or large-scale foundation modification and governance are recommended, such as soil solidification and geotechnical modification. Finally, a zoning map of suitable, restricted, and prohibited construction areas is generated, along with specific governance and design recommendations. All results can be exported through a GIS or BIM system for easy use by planning and design units, and a complete digital report is provided. All output results will include zoning levels, governance measures, design recommendations, and recommended monitoring, forming a conclusion on the suitability of construction in coal mining subsidence areas.

[0059] The specific steps for performing bidirectional complementary correction based on drilling and geophysical data are as follows:

[0060] Step S1: Using geophysical data as input, perform denoising, static correction, amplitude normalization, and gridding on the geophysical data in a unified coordinate system and time base to obtain the preliminary geophysical inversion volume; the preliminary geophysical inversion volume is generated through the inversion process of geophysical data, and each voxel represents the parameter value of the geophysical inversion result, such as resistivity, wave impedance, velocity, etc.

[0061] Step S2: Using the borehole strata, strata elevation, and borehole attributes provided by the drilling data as calibration anchor points, define the comparison window as a depth range with a fixed thickness surrounding the center of the target stratum; compare the preliminary geophysical inversion body with the borehole record segment by segment along the borehole depth to form a time-depth comparison table; adjust the depth positioning of the preliminary geophysical inversion body according to the comparison table to ensure that the preliminary geophysical inversion body and the borehole data are consistent in terms of target strata and attributes; within each comparison window, based on the proportional relationship between the borehole attribute values ​​and the values ​​of the preliminary geophysical inversion body, use the linear scaling method of the scaling factor to recalibrate the voxel amplitude of the window, that is, use the ratio of the borehole attribute value / geophysical value as the amplitude adjustment factor to perform multiplicative correction on the geophysical value, thereby obtaining the calibrated and corrected geophysical inversion body;

[0062] Step S3: Calculate the difference between the calibrated geophysical inversion body and the measured attributes of the borehole at each borehole location to form a residual set. The strike and dip of the coal seam are obtained by comprehensively considering the stratigraphic extension azimuth, geological structure map, and geophysical profile interpretation results from the drilling log. The method is as follows: extract the roof elevation points of the target coal seam from multiple known boreholes, and use least squares plane fitting to obtain the strike and dip. Based on the strike and dip of the coal seam, set the horizontal search radius, vertical search radius, and azimuth angle to establish a search ellipsoid with the strike as the major axis and the dip as the minor axis. Within the search ellipsoid, use the anisotropic inverse distance weighting method to extend the borehole residuals to the surrounding voxels to generate a residual field. Combine this with a preset relaxation coefficient to update the geophysical inversion body to achieve accurate correction.

[0063] Step S4: Scoring of drilling data, calculating scores for core recovery rate, well inclination, and logging completeness, and standardizing them to obtain a score value for each borehole; calculating scores for coverage quality, signal-to-noise ratio, and inversion fitting error for geophysical voxels, and standardizing them to obtain a score value for each voxel; standardizing all scores to the same interval through a unified interval mapping and segmented threshold table.

[0064] Select several calibration boreholes and corresponding geophysical voxels within the project area to form a calibration subset, and perform the update process from step S2 to step S3 once for each of the preset coefficient combinations, calculating the weighted error for each coefficient combination; the weighted error is the square root of the squared difference between the geophysical inversion volume and the measured attribute of each borehole; the coefficient combination with the smallest weighted error is used as the fixed borehole weight coefficient group and geophysical voxel weight coefficient group for this project.

[0065] Using a fixed combination of coefficients, borehole scores are linearly synthesized into total borehole weight, and geophysical scores are linearly synthesized into total geophysical voxel weight. During residual interpolation, "distance attenuation weight × total borehole weight" is used as the comprehensive interpolation weight to participate in anisotropic inverse distance weighting. When updating geophysical voxels, the update magnitude is determined by consulting a graded relaxation coefficient table based on the total geophysical voxel weight. The graded relaxation coefficient table is pre-set by professionals in the relevant field based on historical project experience and numerical simulation results to ensure that the update magnitude is larger in high-weight areas and smaller in low-weight areas.

[0066] Step S5: When there are borehole segments in the drilling data with a core recovery rate lower than the preset core recovery rate threshold and marked as broken, according to the project's preset stratigraphic coding rules (determined based on regional unified stratigraphic naming and geological correlation results), select several high-weight boreholes as reliable boreholes within the same stratigraphic level. Select the two closest reliable boreholes in order of distance along the stratigraphic direction and perform linear interpolation to obtain the preliminary attribute values ​​of the borehole segment. Collect the preliminary attribute values ​​of multiple reliable boreholes within the stratigraphic plane with a preset search radius, and perform a weighted average according to distance attenuation to obtain the smoothed value of the borehole segment. According to the preset weight ratio, synthesize the initial value and the smoothed value into the final attribute completion result of the borehole segment, and assign completion confidence levels to the completion results, which are divided into three levels: high, medium, and low.

[0067] When the borehole trajectory deviates from the plane or depth of the geophysical anomaly center by more than a preset threshold, the borehole trajectory coordinates and sampling depth are corrected by an equal amount according to the deviation direction, and the correction mark is recorded in the borehole data. The corrected coordinates and depth are used in subsequent iterations.

[0068] Step S6: Repeat steps S2 to S5. After each iteration, calculate the weighted root mean square error index and compare it with a preset threshold. The weighted root mean square error index is the square of the difference between the geophysical inversion body and the measured attributes of each borehole. It is calculated by weighting according to the weight of the borehole, and the average value is obtained and the square root is taken. If the weighted root mean square error is less than the preset threshold, or the decrease ratio of the error between two adjacent rounds is less than the preset threshold, or the number of iterations reaches the upper limit, the iteration is terminated.

[0069] After termination, a unified underground 3D model is output. This model continuously represents the goaf boundary, coal pillar geometry, overburden failure zone, and porosity distribution in the same coordinate system, and outputs a residual confidence field. Based on the high confidence areas in the residual confidence field, the geometric center of the connected area is extracted, and the plane coordinates of the borehole are determined. The design depth of the borehole is determined based on the combination of the upper boundary of the goaf, the lower boundary of the goaf, and the lower boundary of the overburden failure zone in the unified underground 3D model, with a safety margin reserved (the safety margin is determined by the project's geological conditions and construction safety specifications, generally not less than 10% of the overburden failure height or a fixed value of 2–5m). A borehole suggestion list is generated, which contains the plane coordinates and design depth of the suggested boreholes. This list is reviewed by qualified geological exploration and drilling engineers, taking into account factors such as on-site construction feasibility, terrain conditions, construction access, and existing engineering layout. If necessary, the coordinates, depths, or suggested borehole positions are adjusted or added / deleted.

[0070] By using the coal pillar size, porosity, and overburden structure information in a unified underground 3D model, and combining the strength, deformation, and permeability parameters from geotechnical test data, the safety factor of the coal pillar, the overburden failure height, and the filling degree are calculated to obtain the foundation stability level. The specific operation is as follows:

[0071] The planar boundaries, column width, column height, planar area, and overburden thickness at the center point of the coal pillar are extracted from the unified underground 3D model; the porosity distribution of voxels in the goaf is extracted and the goaf is divided into blocks according to the goaf connectivity; uniaxial compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, and natural unit weight of rock are extracted from the geotechnical test data, and the roof lithology is divided into hard roof, medium roof, and soft roof according to a preset threshold table, which is set at the start of the project and remains unchanged during the evaluation process;

[0072] The actual load and average stress borne by the coal pillar are calculated using the load area method. Specifically, the load area of ​​the coal pillar is defined by drawing a perpendicular line from the plane boundary of the coal pillar to the boundary of the adjacent roadway or goaf. The product of the overburden thickness and the natural density of the rock is taken as the vertical overburden stress, which is multiplied by the load area to obtain the load on the coal pillar. The average stress of the coal pillar is then calculated using the plane area of ​​the coal pillar. The ultimate bearing capacity of the coal pillar is calculated using the pillar strength reduction method. First, the uniaxial compressive strength is used as the base value, and then it is corrected according to the reduction coefficient corresponding to the pillar width-to-height ratio, the degree of joint development, and the saturation state to obtain the effective pillar strength. The effective pillar strength is multiplied by the plane area of ​​the coal pillar to obtain the ultimate bearing capacity of the coal pillar. The ratio of the ultimate bearing capacity of the coal pillar to the actual load of the coal pillar is calculated to obtain the safety factor of the coal pillar.

[0073] Based on the mining thickness or equivalent mining thickness provided by the unified underground 3D model, the roof failure coefficient is selected from the preset coefficient table according to the roof lithology category. The product of the roof failure coefficient and the mining thickness is calculated to obtain the overburden failure height. When the fault and bedrock aquifer represented by the unified underground 3D model appear within a certain buffer distance above the coal pillar, the overburden failure height is incrementally corrected according to the preset correction coefficient. The overburden failure height is then output.

[0074] For each goaf block, the porosity is weighted and averaged using voxel volume as the weight to obtain the block average porosity; the goaf filling degree is calculated according to the formula: filling degree = 1 − block average porosity, and the filling degree field and block-level statistical values ​​are marked in the 3D model; when there are multiple coal seams stacked, the filling degree is first calculated layer by layer according to the vertical juxtaposition relationship of the blocks, and then equivalent synthesis is performed using layer thickness as the weight.

[0075] The coal pillar safety factor, overburden failure height, and filling degree are used as the control parameter set, and stability is classified according to the preset classification rules: when the coal pillar safety factor is not lower than the first threshold, the overburden failure height does not exceed the roof safety limit, and the filling degree is not lower than the first threshold, it is judged as stable; when any indicator falls into the second threshold range, it is judged as basically stable; all other cases are judged as unstable; the foundation stability level map, coal pillar safety factor map, overburden failure height map, and filling degree map are output.

[0076] The above calculation steps and formulas, such as the load area method, the column strength reduction method, the overburden failure height calculation formula, and the filling degree calculation formula, are all common knowledge and standard calculation methods in the field of engineering geology, and are widely used in stability analysis of coal mining and underground engineering. Specifically:

[0077] The load area method is used to calculate the load and average stress on a coal pillar. It is a commonly used coal pillar load distribution method in the engineering field. The bearing capacity of the coal pillar is determined by calculating the load area of ​​the coal pillar and the stress distribution of the overlying strata.

[0078] The column strength reduction method is used to estimate the ultimate bearing capacity of a coal column. It uses uniaxial compressive strength as the benchmark and reduces it according to the geometry, joint development, and saturation state of the coal column. This method is widely used in geotechnical engineering and conforms to industry standards.

[0079] The formula for calculating the overburden failure height uses the relationship between the roof lithology and the mining thickness to estimate the height of the overburden failure. When special geological conditions (such as faults and aquifers) occur, a correction factor is introduced. It is usually used to assess the stability of overburden in coal mine engineering.

[0080] The filling degree is calculated based on the weighted average of the porosity of the goaf, using the standard formula: Filling degree = 1 − average porosity of the block. This method has been widely used in geological exploration and mining engineering to assess the filling status of goafs. The above methods and formulas are known technologies and do not involve any innovative settings of this invention.

[0081] The feature influence weights are calculated based on physical sensitivity, SHAP contribution, and data confidence. The specific steps are as follows:

[0082] Under a unified coordinate and time datum, a feature set is constructed for each monitoring point and within a fixed time window. The feature set consists of the following data: displacement, velocity, and acceleration time series quantities from the settlement monitoring sequence; static attributes provided by the unified underground 3D model, including coal pillar size, porosity, overburden failure zone markers, equivalent mining thickness, and geometric quantities from the goaf to the monitoring point; strength, deformation, and permeability parameters from geotechnical test data; elevation and slope from remote sensing topographic data; and mining depth, mining thickness, and mining period from historical data.

[0083] Using the improved probability integral method as the baseline model, the cumulative settlement index of the baseline is calculated within the time window; for each feature in the feature set, a positive perturbation and a negative perturbation are performed according to the preset perturbation table, and the change in the cumulative settlement index is calculated respectively. The larger of the two absolute values ​​of the change is taken as the original value of the physical sensitivity of the feature; the original value is standardized to [0,1] according to the unified interval mapping table to obtain the set of physical sensitivity scores.

[0084] A gradient boosting tree surrogate model with a fixed structure is trained on the aligned feature set. The input is the feature set, and the output is the cumulative settlement index within the time window. Based on the trained surrogate model, TreeSHAP is used to calculate the mean absolute SHAP value of each feature on the training set and the validation set, and the value is standardized to [0,1] according to a unified interval mapping table to obtain the SHAP contribution score set.

[0085] Coverage score, noise score, and stability score are calculated for time-series elements and standardized to [0,1]. Voxel total weight score and residual confidence score are calculated for static elements and standardized to [0,1]. The three time-series scores are combined into a time-series confidence score according to a fixed coefficient table, and the two static scores are combined into a static confidence score. The time-series confidence score and the static confidence score are then combined into a data confidence score set according to the fixed coefficient table, which is used as the output. The coefficient table is determined at the start of the project and remains unchanged during the project cycle.

[0086] For each feature, its physical sensitivity score, SHAP contribution score, and data confidence score are taken and combined into an influence weight of that feature using a fixed coefficient combination. The coefficient combination is determined at the start of the project and fixed during the project cycle. The influence weight is directly used in two places in the neural network: first, input-side scaling, that is, scaling each channel of the feature set according to the corresponding influence weight before feeding it into the network; second, sample loss weighting, that is, using the weighted mean of the influence weights of each feature as the loss weight of the sample to weight the training loss.

[0087] The residual deformation prediction model is based on an LSTM model and includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer. The input layer receives the settlement monitoring sequence and static attributes and normalizes them. The LSTM layer processes the time series data and captures the long-term dependencies of the time series information. The attention mechanism layer calculates a weighted sum based on the influence weight of each feature, enabling the model to automatically focus on features that have a greater impact on residual deformation prediction. The fully connected layer performs nonlinear mapping on the features processed by the LSTM layer and the attention mechanism layer and performs feature dimensionality reduction. The output layer generates predicted values ​​for residual settlement, tilt, and horizontal deformation in future time periods.

[0088] The specific steps for training the residual deformation prediction model are as follows:

[0089] Obtain several labeled training samples. Each training sample contains time-series data extracted from settlement monitoring sequences, static attribute data extracted from a unified underground 3D model, and corresponding labels for actual residual settlement, tilt, and horizontal deformation. Randomly divide all training samples into training and validation sets, ensuring an appropriate ratio between the two sets to avoid overfitting. Initialize the parameters of the residual deformation prediction model and iteratively train the model using the training set. After each training round, use the validation set as input to calculate the validation accuracy and loss value. Set termination conditions, including the validation accuracy reaching a preset threshold or no significant decrease in validation loss for several consecutive rounds. When the validation result meets any termination condition, output the trained residual deformation prediction model. If the validation result does not meet the termination condition, continue iteratively updating the residual deformation prediction model using the training set until the termination condition is met.

[0090] The influence depth is calculated based on building loads from historical data or design parameters, and coupled with the overburden failure height and minimum mining depth given by a unified underground 3D model to determine the load risk level. The specific operation is as follows:

[0091] Based on the building load data provided in historical records, the depth of influence of building load on the foundation is calculated. Specifically, based on the load type, design height, and foundation area of ​​the building, the additional stress on the underground structure under the load is calculated using the existing foundation stress propagation theory. According to the relationship between the additional stress and the self-weight stress of the foundation, the depth of influence is calculated, that is, the maximum range of the depth of influence when the ratio of additional stress to self-weight stress reaches a certain standard.

[0092] Based on the calculated overburden failure height and the minimum mining depth given by the unified underground 3D model, it is determined whether the area is suitable for construction. The load influence depth, overburden failure height, and minimum mining depth are coupled for judgment: if the minimum mining depth is greater than the load influence depth plus the overburden failure height, the area is judged as a suitable construction area; if the minimum mining depth is close to the load influence depth plus the overburden failure height, the area is a restricted construction area, and reinforcement measures are recommended; if the minimum mining depth is less than the load influence depth plus the overburden failure height, the area is judged as a prohibited construction area, and construction prohibition measures must be taken.

[0093] Based on the coupled judgment results of load influence depth, overburden failure height, and minimum mining depth, combined with the foundation stability and residual deformation assessment results, the load risk level is output. The risk level is divided into three levels: low risk, medium risk, and high risk. The low risk zone indicates that the foundation can withstand the building load, has good stability, and is suitable for construction. The medium risk zone indicates that the foundation may have a certain deformation risk, is suitable for construction but requires reinforcement measures. The high risk zone indicates that the foundation is unstable, is not suitable for construction, and requires prohibition of construction or other remediation measures.

[0094] Example 2: An intelligent analysis system for geological exploration data in coal mining subsidence areas, such as... Figure 1 As shown, it includes:

[0095] The data acquisition module is used to acquire drilling data, geophysical data, remote sensing topographic data, monitoring and measured data, geotechnical test data, and historical data of the target coal mining subsidence area. The monitoring and measured data includes GNSS displacement time series, leveling displacement time series, and InSAR displacement time series, which are defined as settlement monitoring sequences. The remote sensing topographic data provides topographic parameters, and the historical data provides mining depth, mining thickness, and building load parameters.

[0096] The underground 3D model construction module is used to perform bidirectional complementary correction between drilling and geophysical data. Specifically, it includes: calibrating the initial geophysical inversion volume using borehole strata and amplitude as anchor points, and performing iterative correction by establishing point-surface residuals; simultaneously, completing the attributes of geophysical voxels with low core recovery rates and broken blind sections; and outputting a unified underground 3D model after confidence-weighted convergence, including goaf boundaries, coal pillar geometry, overburden failure zones, continuous porosity distribution, and residual confidence fields, and generating a list of borehole repair suggestions.

[0097] The foundation stability assessment module is used to calculate the safety factor of the coal pillar, the overburden failure height and the filling degree, etc., based on the coal pillar size, porosity and overburden structure information in the unified underground three-dimensional model, combined with the strength, deformation and permeability parameters in the geotechnical test data, to obtain the foundation stability level, and to provide constraints for subsequent residual deformation prediction and load judgment.

[0098] The residual deformation prediction module is used to take the settlement monitoring sequence as the time-series input and the static attributes extracted from the unified underground 3D model, geotechnical test, remote sensing topography and historical data as auxiliary input. The feature influence weight is calculated by physical sensitivity, SHAP contribution and data confidence. The residual deformation prediction model outputs the subsidence, tilt and horizontal deformation fields for future periods.

[0099] The suitability assessment module is used to calculate the impact depth of building loads based on historical data or design parameters, and to couple the assessment with the overburden failure height and minimum mining depth given by the unified underground 3D model to obtain the load risk level. Finally, the foundation stability level, residual deformation results and load risk are input into the rule engine to automatically generate suitable, restricted and prohibited construction zones and supporting treatment and design suggestions, thereby forming a conclusion on the suitability of construction in the coal mining subsidence area.

[0100] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for intelligent analysis of geological exploration data of a coal mining subsidence area, characterized in that, include: The drilling data, geophysical data, remote sensing topographic data, monitoring and measured data, geotechnical test data, and historical data of the target coal mining subsidence area are acquired. The monitoring and measured data includes GNSS displacement time series, leveling displacement time series, and InSAR displacement time series, and the above displacement time series are defined as settlement monitoring series. The remote sensing topographic data provides topographic parameters, and the historical data provides mining depth, mining thickness, and building load parameters. Two-way complementary correction is performed based on drilling and geophysical data: the initial geophysical inversion body is anchored by the borehole strata and amplitude, point-to-surface residual iterative correction is established, and attribute completion is performed on voxels with low core recovery rate and broken blind sections; convergence is performed by confidence weight, and a unified underground 3D model is output, including goaf boundary, coal pillar geometry, overburden failure zone, porosity distribution and residual confidence field, and borehole repair suggestions are generated; By using the coal pillar size, porosity and overburden structure information in the unified underground three-dimensional model, and combining the strength, deformation and permeability parameters in the geotechnical test data, the safety factor of the coal pillar, the overburden failure height and the filling degree are calculated to obtain the foundation stability level. Using the settlement monitoring sequence as the time-series input and the static attributes extracted from the unified underground 3D model, geotechnical test, remote sensing topography and historical data as auxiliary input, the feature influence weights are calculated by physical sensitivity, SHAP contribution and data confidence, and the subsidence, tilt and horizontal deformation fields of future periods are output through the residual deformation prediction model. The influence depth of building load is calculated based on historical data or design parameters, and coupled with the overburden failure height and minimum mining depth given by the unified underground three-dimensional model to obtain the load risk level. The foundation stability level, residual deformation results and load risk are input into the rule engine to automatically generate suitable construction, restricted construction, and prohibited construction zones, as well as supporting treatment and design suggestions, to form a conclusion on the suitability of construction in the coal mining subsidence area.

2. The intelligent analysis method for geological exploration data of coal mining subsidence area according to claim 1, characterized in that, The specific steps for performing bidirectional complementary correction based on drilling and geophysical data are as follows: Step S1: Using geophysical data as input, perform denoising, static correction, amplitude normalization, and gridding on a unified coordinate system and time base to generate a preliminary geophysical inversion body; Step S2: Using borehole strata, elevation, and borehole properties as anchor points, establish a time-depth comparison table segment by segment along the borehole depth to correct the depth positioning of the inverted body; within the comparison window, linearly scale and recalibrate the voxel amplitude of the window according to the ratio of borehole property values ​​to geophysical values ​​to obtain the calibrated geophysical inversion body; Step S3: Calculate the difference between the calibrated inversion body and the measured attributes at each borehole location to form a residual set; obtain the strike and dip of the coal seam based on the existing borehole elevation points and interpretation results, and construct an anisotropic search ellipsoid with strike as the major axis and dip as the minor axis; use anisotropic inverse distance weighted diffusion residuals within the ellipsoid to obtain the residual field, and update the geophysical inversion body in combination with preset relaxation coefficients; Step S4: Scoring of drilling data, calculating scores for core recovery rate, well inclination, and logging completeness, and standardizing them to obtain a score value for each borehole; calculating scores for coverage quality, signal-to-noise ratio, and inversion fitting error for geophysical voxels, and standardizing them to obtain a score value for each voxel; standardizing all scores to the same interval through a unified interval mapping and segmented threshold table. Select several calibration boreholes and corresponding geophysical voxels within the project area to form a calibration subset. Iterate through the preset coefficient combinations on the calibration subset and execute steps S2 to S3 to calculate the weighted error. Select the borehole weight and voxel weight corresponding to the minimum error as a fixed combination. When interpolating the residuals, the product of the distance attenuation weight and the total borehole weight is used as the comprehensive weight. The voxel update range is determined by the total voxel weight according to the graded relaxation coefficient table. Step S5: When there are borehole segments in the drilling data with a core recovery rate lower than the preset core recovery rate threshold and marked as broken, select high-weight reliable boreholes in the same layer according to the preset layer coding rules of the project. First, use the linear interpolation of the two nearest boreholes as the initial value, and then perform distance weighting to obtain a smooth value in the layer plane according to the search radius. Then, merge the results into the final complete attribute according to the preset ratio and assign high, medium and low confidence levels. When the borehole trajectory deviates from the plane or depth of the geophysical anomaly center by more than a threshold, the coordinates and sampling depth are corrected by an equal amount according to the direction of deviation and recorded. Step S6: Repeat steps S2 to S5. After each iteration, calculate the weighted root mean square error index and compare it with a preset threshold. The weighted root mean square error index is the square of the difference between the geophysical inversion body and the measured attributes of each borehole. It is calculated by weighting according to the weight of the borehole, and the average value is obtained and the square root is taken. If the weighted root mean square error is less than the preset threshold, or the decrease ratio of the error between two adjacent rounds is less than the preset threshold, or the number of iterations reaches the upper limit, the iteration is terminated. After termination, a unified underground 3D model is output, and a residual confidence field is output. Based on the high confidence region in the residual confidence field, the geometric center of the connected region is extracted, the plane coordinates of the fill hole are determined, a safety margin is reserved, and a list of fill hole suggestions is generated.

3. The intelligent analysis method for geological exploration data in coal mining subsidence areas according to claim 2, characterized in that, By using the coal pillar size, porosity, and overburden structure information in a unified underground 3D model, and combining the strength, deformation, and permeability parameters from geotechnical test data, the safety factor of the coal pillar, the overburden failure height, and the filling degree are calculated to obtain the foundation stability level. The specific operation is as follows: Extract the planar boundary, column width, column height, and overburden thickness at the center point of the coal pillar from the 3D model; extract the porosity distribution of the goaf and divide it into blocks according to connectivity; extract uniaxial compressive strength, elastic modulus, Poisson's ratio, cohesion, internal friction angle, and natural unit weight of rock from the geotechnical test data, and classify the roof lithology into hard, medium, and relatively soft roofs according to a preset threshold table; The load and average stress of the coal pillar are calculated using the load area method. Specifically, the load area of ​​the coal pillar is delineated by drawing a perpendicular line between the plane boundary of the coal pillar and the boundary of the adjacent roadway or goaf. The product of the overburden thickness and the natural unit weight of the rock is taken as the vertical overburden stress, which is then multiplied by the load area to obtain the load on the coal pillar. The average stress of the coal pillar is obtained from the plane area of ​​the coal pillar. The pillar strength reduction method is used to correct the uniaxial compressive strength according to factors such as the pillar width-to-height ratio, calculate the ultimate bearing capacity of the coal pillar, and calculate the ratio of the actual load to the ultimate bearing capacity of the coal pillar to obtain the safety factor of the coal pillar. The overburden failure height is calculated based on the roof lithology and mining thickness; the goaf filling degree is calculated by weighted average of porosity, and when multiple coal seams exist, the calculation is performed layer by layer and weighted composite. Stability is determined according to preset classification rules: when the safety factor of the coal pillar, the overburden failure height and the filling degree all meet the stability conditions, it is determined to be stable; otherwise, it is basically stable or unstable, and the corresponding stability level is output.

4. The intelligent analysis method for geological exploration data in coal mining subsidence areas according to claim 3, characterized in that, The feature influence weights are calculated based on physical sensitivity, SHAP contribution, and data confidence. The specific steps are as follows: Under a unified coordinate and time datum, a feature set is constructed for each monitoring point and fixed time window, including settlement monitoring sequence, static attributes of three-dimensional model, geotechnical test data, remote sensing topographic data and historical data; Using the improved probability integral method as the baseline model, the cumulative settlement index of the baseline is calculated; positive and negative perturbations are applied to each feature, and the change in the cumulative settlement index of the baseline before and after the perturbation is calculated. The absolute values ​​of the changes in the positive and negative perturbations are compared, and the larger absolute value is taken as the original value of the physical sensitivity of the feature. The sensitivity score is obtained by standardizing to [0,1]. Train a gradient boosting tree surrogate model, calculate the SHAP value for each feature based on the model, and obtain the SHAP contribution score after standardization; Coverage, noise, and stability scores are calculated for time-series elements, while voxel total weights and residual confidence scores are calculated for static elements. After standardization, time-series and static confidence scores are synthesized separately. Finally, based on a fixed coefficient combination, the physical sensitivity score, SHAP contribution score, and data confidence score are combined to form the influence weight of each feature.

5. The intelligent analysis method for geological exploration data in coal mining subsidence areas according to claim 4, characterized in that, The residual deformation prediction model is based on the LSTM model and includes an input layer, an LSTM layer, an attention mechanism layer, a fully connected layer, and an output layer. The input layer is used to receive the settlement monitoring sequence and static attributes and normalize them. The LSTM layer is used to process time series data and capture the long-term dependence of time series information. The attention mechanism layer is used to calculate a weighted sum based on the influence weight of each feature to enhance attention to key features for residual deformation prediction; the fully connected layer performs nonlinear mapping on the features processed by the LSTM layer and the attention mechanism layer and performs feature dimensionality reduction; the output layer is used to generate predicted values ​​of residual settlement, tilt and horizontal deformation for future time periods.

6. The intelligent analysis method for geological exploration data in coal mining subsidence areas according to claim 5, characterized in that, The influence depth of building loads is calculated based on historical data or design parameters, and coupled with the overburden failure height and minimum mining depth given by a unified underground 3D model to determine the load risk level. The specific operation is as follows: Based on the building load type, design height, and foundation area, the additional stress on the underground structure is calculated using the foundation stress propagation theory, and the load influence depth is calculated based on the ratio of additional stress to self-weight stress. Based on the overburden failure height and minimum mining depth, determine the regional construction conditions: if the minimum mining depth is greater than the load influence depth plus the overburden failure height, it is a suitable construction area; if it is close to it, it is a restricted construction area, and reinforcement is recommended; if it is less than it, it is a prohibited construction area, and construction is prohibited or remediation is required. Based on the coupling results of load influence depth, overburden failure height and minimum mining depth, combined with foundation stability and residual deformation assessment, the load risk level is output, which is divided into low risk, medium risk and high risk.

7. An intelligent analysis system for geological exploration data in coal mining subsidence areas, characterized in that, The system is applied to the intelligent analysis method for geological exploration data in coal mining subsidence areas as described in any one of claims 1-6, including: The data acquisition module is used to acquire drilling data, geophysical data, remote sensing topographic data, monitoring and measurement data, geotechnical test data and historical data of the target coal mining subsidence area; The underground 3D model building module is used to perform two-way complementary correction between drilling and geophysical data, output a unified underground 3D model, and generate a list of borehole repair suggestions. The foundation stability assessment module is used to calculate indicators such as the coal pillar safety factor, overburden failure height and filling degree based on the coal pillar size, porosity and overburden structure information in the unified underground three-dimensional model, combined with the strength, deformation and permeability parameters in the geotechnical test data, to obtain the foundation stability level. The residual deformation prediction module is used to output the subsidence, tilt, and horizontal deformation fields for future periods through the residual deformation prediction model. The suitability assessment module is used to obtain the load risk level, generate suitable, restricted, and prohibited construction zones, and provide supporting governance and design suggestions to form a conclusion on the suitability of construction in coal mining subsidence areas.

Images