Intelligent analysis method for geomechanical conditions of coal seam surrounding rock

Abstract

The invention discloses an intelligent analysis method for geomechanical conditions of coal seam surrounding rocks, and belongs to the field of intelligent analysis of the geomechanical conditions of the coal seam surrounding rocks. The intelligent analysis method comprises the following steps: performing time sequence matching processing on historical geological data and real-time monitoring information by adopting a dynamic data integration technology according to a preliminary integration result of multi-source data; determining a unified feature set after multi-source data fusion; acquiring geomechanical condition related data of the unmined area according to the initial prediction model parameters, and performing simulation analysis on the surrounding rock stability to obtain stability prediction distribution of the unmined area; according to the stability prediction distribution of the unmined area, combining a dynamic data integration result, carrying out local correction on the prediction distribution, and determining a final stability prediction result; and for a final stability prediction result, generating a geomechanical condition risk distribution diagram of the unmined area through an intelligent prediction technology, obtaining distribution characteristics of a potential risk area, and completing output of a prediction model.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent analysis of geomechanical conditions of coal seam surrounding rocks, and in particular relates to an intelligent analysis method of geomechanical conditions of coal seam surrounding rocks. Background Art

[0002] Intelligent analysis of the geomechanical conditions of coal seam surrounding rocks is a key research area to ensure safe production and efficient mining in coal mines. The stability of the surrounding rock directly affects the design of tunnel support and the prevention and control of geological disasters, and accurate prediction of the mechanical behavior of the surrounding rock is crucial to reducing the risk of accidents. Currently, most analysis methods rely on traditional geological exploration methods and empirical formulas, and have problems with insufficient data integration and limited analysis depth. These methods often have difficulty in effectively processing multi-source heterogeneous data, such as the complex relationship between stratum lithology, tectonic stress and hydrogeological information, resulting in inaccurate prediction results in complex geological environments. In addition, when dealing with the distribution characteristics of the fracture network inside the surrounding rock, the existing technology lacks the ability to quantify the sparsity characteristics, making it difficult to accurately characterize the heterogeneity of the surrounding rock structure.

[0003] The core challenge facing the research lies in how to efficiently characterize the sparsity of the surrounding rock fracture network and, on this basis, achieve dynamic fusion and prediction of multi-source data. The sparsity of the fracture network is a key factor affecting the stability of the surrounding rock, but its distribution pattern is complex and difficult to accurately identify through traditional means. This makes it difficult to establish a correlation model between fracture distribution and surrounding rock stability. Furthermore, the quantification of sparsity characteristics needs to be combined with the integration of multi-source geological data, and existing methods have difficulty in dynamically fusing historical geological information with real-time monitoring data, resulting in insufficient prediction accuracy for geomechanical conditions in unmined areas. These two factors are interrelated: inaccurate sparsity characterization will weaken the effect of multi-source data fusion, and the lack of data fusion, in turn, limits the in-depth analysis of sparsity characteristics.

[0004] Therefore, how to develop a fracture network identification technology based on sparsity quantification and combine it with a dynamic fusion method of multi-source data to establish an intelligent prediction model for geomechanical conditions in unmined areas has become a key issue in this study. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides an intelligent analysis method for the geomechanical conditions of coal seam surrounding rocks, comprising:

[0006] Digitally process image data and geological exploration records of the surrounding rock fracture network to obtain a preliminary characteristic dataset of fracture distribution;

[0007] Based on the preliminary feature data set, a sparsity quantitative analysis method is used to deeply mine the crack distribution characteristics and construct a sparsity feature vector;

[0008] By combining the correlation between the sparsity characteristic vector and the heterogeneity of the surrounding rock structure, the stratigraphic lithology and tectonic stress information in the historical geological data, as well as the stress changes and hydrogeological parameters in the real-time monitoring information, are obtained to obtain the preliminary integration results of the multi-source data;

[0009] Performing time series matching processing on the historical geological data and real-time monitoring information in the preliminary integration results by using dynamic data integration technology to determine a unified feature set after multi-source data fusion;

[0010] Based on the unified feature set, a feature mapping relationship of geomechanical conditions is constructed, and combined with the sparsity feature vector, whether the input parameters of the surrounding rock stability prediction model meet the preset threshold requirements is determined, and if so, initial prediction model parameters are generated;

[0011] Obtain relevant data on geomechanical conditions in unmined areas, simulate and analyze surrounding rock stability using intelligent prediction technology, and obtain predicted stability distribution in unmined areas;

[0012] If abnormal stress changes are detected in the real-time monitoring information, the stability prediction distribution is locally corrected to determine the final stability prediction result;

[0013] Based on the final stability prediction results, a geomechanical condition risk distribution map of the unmined area is generated, the distribution characteristics of the potential risk area are obtained, and the output of the prediction model is completed.

[0014] Preferably, the step of digitally processing the image data of the surrounding rock fracture network and the geological exploration records to obtain a preliminary characteristic data set of fracture distribution includes:

[0015] Process network images and geological exploration data of surrounding rock fractures to obtain digitally converted raw data sets;

[0016] Image processing technology is used to perform denoising and edge detection on the network image, extract preliminary contour information of crack distribution, and obtain an initial feature set of crack distribution.

[0017] Preferably, the step of deeply mining the crack distribution characteristics using a sparsity quantitative analysis method based on the preliminary feature data set and constructing a sparsity feature vector comprises:

[0018] Obtaining initial distribution information from the preliminary feature data set, and using a data preprocessing method to clean and format the raw data of the fracture network to obtain a preliminary fracture distribution data set;

[0019] Based on the preliminary fracture distribution data set, a sparse characteristic analysis method is used to extract sparsity-related parameters and determine a sparse characteristic index in the fracture network;

[0020] According to the sparse characteristic index, a quantitative analysis technology is used to construct a feature vector, obtain the sparsity feature vector expression of the fracture network, and construct a sparsity feature vector.

[0021] Preferably, the step of combining the association between the sparsity characteristic vector and the surrounding rock structural heterogeneity, acquiring stratum lithology and tectonic stress information in historical geological data, and stress changes and hydrogeological parameters in real-time monitoring information, and obtaining a preliminary integration result of multi-source data includes:

[0022] Obtain stratum lithology and tectonic stress information from historical geological data, and obtain a structured geological information dataset through data cleaning and format standardization;

[0023] Based on the stress changes and hydrogeological parameters in the real-time monitoring information, a time series analysis method is used to extract the dynamic characteristics of stress changes and the fluctuation laws of hydrogeology, and determine the key change trends of the real-time data;

[0024] Based on the key change trends of the real-time data and in combination with the structured geological information dataset, a mapping relationship between the sparsity feature vector and the surrounding rock structural heterogeneity is constructed to determine the potential correlation between the two;

[0025] If the correlation strength in the mapping relationship is lower than a preset threshold, the historical geological data and the real-time monitoring data are weightedly integrated through multi-source data fusion technology to obtain a preliminary integration of the multi-source data.

[0026] Preferably, the step of performing time series matching processing on the historical geological data and real-time monitoring information in the preliminary integration result by using dynamic data integration technology to determine a unified feature set after multi-source data fusion includes:

[0027] Using a time series analysis method to match the historical geological data and real-time monitoring information in the preliminary integration results to construct a unified feature set;

[0028] For the unified feature set, an interpolation method is used to align the data to obtain an aligned feature data set;

[0029] For the aligned feature data set, the distribution characteristics of geological information and monitoring information are obtained, and the correlation strength between the data is determined. If the correlation strength is lower than a preset threshold, the data is subjected to secondary fusion processing to determine a unified feature set after multi-source data fusion.

[0030] Preferably, the process of generating the initial prediction model parameters includes:

[0031] Using a data integration tool to standardize geological information from different sources in the unified feature set to obtain a standardized feature data set;

[0032] Based on the normalized feature data set, combining geomechanics with feature mapping, constructing a feature correlation matrix of mechanical conditions to determine dependency relationships between features;

[0033] Based on the dependency relationship between the features and the sparse feature vectors, a sparse representation method is applied to reduce the dimension of the feature matrix to obtain a simplified feature vector set;

[0034] According to the simplified feature vector set, the feature vector is matched by a parameter screening tool to determine whether it meets the preset threshold requirements. If so, a preliminary model parameter set is generated;

[0035] A support vector machine algorithm is used to optimize and adjust the preliminary model parameter set to obtain an adjusted model parameter combination;

[0036] A verification process is performed based on the adjusted model parameter combination. If the verification result meets the preset stability standard, the final prediction model parameters are determined.

[0037] Preferably, the process of determining the final stability prediction result includes:

[0038] Obtain real-time monitoring data from monitoring equipment, use time series analysis methods to determine whether there are abnormal stress changes, and obtain stress anomaly data;

[0039] If the stress abnormal data is detected, the abnormality degree is judged by a preset threshold, and the K-means clustering algorithm is used to classify the abnormal data to determine the range of the abnormal area;

[0040] According to the range of the abnormal area, dynamic data of the unmined area is obtained from the database, and the dynamic data and real-time monitoring data are integrated using data fusion technology to obtain a comprehensive data set;

[0041] For the comprehensive data set, a support vector machine algorithm is used to analyze the stability prediction distribution of the unmined area to obtain an initial stability prediction distribution;

[0042] If there is an abnormal area in the initial stability prediction distribution, the abnormal area is locally corrected, and the prediction distribution parameters are adjusted using a weighted average method to obtain a corrected prediction distribution;

[0043] The final stability prediction result of the unmined area is determined by using the modified prediction distribution and the probability density estimation method.

[0044] Preferably, the step of completing the output of the prediction model includes:

[0045] Extract geomechanical parameters from existing data, construct an initial data set, and use the support vector machine algorithm to classify the data to obtain preliminary geomechanical condition classification results.

[0046] Based on the classification results, a risk assessment matrix is ​​constructed for the geomechanical conditions of the unmined area to obtain preliminary distribution characteristics of potential risks and determine an initial mapping of risk level distribution;

[0047] By performing spatial interpolation processing on the initial map, a risk distribution map of the unmined area is generated to obtain more detailed regional distribution characteristics and determine the boundary range of the potential risk area;

[0048] If the risk level of certain areas in the risk distribution map exceeds a preset threshold, the geomechanical data of these areas are deeply analyzed to obtain the specific distribution locations of high-risk areas;

[0049] Based on the distribution of the high-risk areas and combined with stability prediction data, targeted risk identification results are generated to determine the scope of key monitoring areas;

[0050] By integrating the scope division of the key monitoring areas with the output of the prediction model, the final risk distribution characteristic data is generated to obtain complete potential risk area distribution information;

[0051] If there are missing data or abnormal points in the final distribution information, the missing parts are supplemented by the spatial neighborhood analysis method to obtain complete risk distribution map data.

[0052] On the other hand, the present invention further provides an electronic device, comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the method is implemented when the processor executes the computing program.

[0053] On the other hand, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the method when executed by a processor.

[0054] Compared with the prior art, the present invention has the following advantages and technical effects:

[0055] The present invention discloses a method for predicting surrounding rock stability based on the sparsity characteristics of the fracture network. By digitally processing the surrounding rock fracture network image data and geological exploration records, the sparsity characteristics of the fracture distribution are extracted, and the historical geological data and real-time monitoring information are combined to construct a unified feature set of multi-source data fusion. On this basis, the present invention establishes a mapping relationship of geomechanical condition characteristics, uses intelligent prediction technology to simulate and analyze the stability of the surrounding rock in unmined areas, and dynamically corrects the prediction results based on real-time monitoring information, and finally generates a geomechanical condition risk distribution map for the unmined area. Through fracture network sparsity analysis and multi-source data fusion, the present invention realizes the accurate prediction of surrounding rock stability under complex geological conditions, providing important decision-making support for mine safety production. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0057] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0059] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0060] Example 1

[0061] like Figure 1 As shown, this embodiment provides an intelligent analysis method for the geomechanical conditions of the surrounding rock of a coal seam, including:

[0062] Step S101 , by digitally processing the image data of the surrounding rock fracture network and the geological exploration records, a preliminary characteristic data set of fracture distribution is obtained, and the sparsity in the fracture distribution characteristics is preliminarily extracted to obtain a basic description of the sparsity of the fracture network.

[0063] By processing network images of surrounding rock fractures and geological exploration data, a digitally converted raw data set is obtained. Image processing techniques are used to denoise and detect edges in the network images, extracting preliminary outline information of the fracture distribution and obtaining an initial feature set for the fracture distribution. This initial feature set is then screened for sparsity characteristics. Data points in the initial feature set are classified using a preset threshold. Data points with a distribution density below the preset threshold are classified as exhibiting sparsity characteristics, resulting in a sparsity feature subset. Spatial analysis of the sparsity feature subset is performed to determine the sparsity distribution pattern of the fracture network. Cluster analysis is used to group the data points in the sparsity feature subset, identifying the spatial clustering pattern of the sparsity features and obtaining a structured description of the sparsity distribution. Based on this structured description of the sparsity distribution, key sparsity parameters of the fracture network are extracted. Based on the clustering pattern in the structured description, the sparsity index of each region is calculated to obtain a sparsity parameter set. Feature mapping of the sparsity parameter set is performed to obtain a quantitative expression of the sparsity of the fracture network. Principal component analysis (PCA) was used to reduce the dimensionality of the sparsity parameter set, extract the main eigenvectors, and determine the quantitative results of the sparsity characteristics. Based on the quantified sparsity characteristics, a basic description of the fracture network was generated. The main eigenvectors in the quantified results were corrected with geological exploration data to obtain the final description of the fracture network sparsity.

[0064] Specifically, by digitizing the image data and geological exploration records of the surrounding rock fracture network, a high-resolution camera was first used to obtain fracture image data with a resolution set to 0.1 mm per pixel. The image data was then spatially aligned with the recorded data, combined with the fracture location and depth data in the geological exploration records. A feature point-based image registration algorithm, such as the SIFT algorithm, was used to extract the fracture edge feature points in the image and match them with the coordinate points in the exploration records. The matching accuracy was controlled within 0.5 mm to generate a unified digital fracture distribution data set. Subsequently, a preliminary extraction of the sparsity in the fracture distribution characteristics was performed, and spatial statistical analysis methods, such as K-function analysis, were used to calculate the spatial aggregation between fracture points. The analysis radius was set to 0.2 meters, and the aggregation index of the fracture distribution was obtained to be 1.8, indicating that the fracture distribution has certain sparsity characteristics. At the same time, combined with the fracture density calculation, the number of fractures per square meter was 3.5, further quantifying the sparsity. Finally, based on the above analysis results, a basic description of the sparsity of the fracture network was constructed. Using the sparse matrix representation, the fracture network was simplified into a two-dimensional sparse matrix with a matrix dimension of 100x100, in which the non-zero elements accounted for only 2.3%, reflecting the sparse characteristics of the fracture distribution. A sparsity distribution map was generated using a visualization tool, which intuitively showed that the sparse fracture area accounted for 78%, providing data support for subsequent fracture network modeling.

[0065] Step S102: Based on the basic description of the sparsity of the fracture network, a sparsity quantitative analysis method is used to deeply explore the fracture distribution characteristics, construct a sparsity feature vector, and determine the quantitative expression of the fracture network in the surrounding rock structure heterogeneity.

[0066] By obtaining initial distribution information from the fracture network data and using data preprocessing methods to clean and format the raw data of the fracture network, a preliminary fracture distribution dataset was obtained. Based on the preliminary fracture distribution dataset, a sparse characteristic analysis method was used to extract sparsity-related parameters and determine the sparse characteristic indicators in the fracture network. Based on the sparse characteristic indicators, a feature vector was constructed using quantitative analysis technology to obtain the sparsity characteristic vector expression of the fracture network. By deeply mining the sparsity characteristic vector, the correlation between the fracture distribution and the surrounding rock structure was analyzed to determine the heterogeneous distribution pattern. If the heterogeneous distribution pattern meets the preset threshold conditions, the support vector machine algorithm is used to classify the heterogeneity of the fracture distribution and the surrounding rock structure to obtain the classified structural characteristic results. Based on the classified structural characteristic results, a quantitative expression model of the fracture network in the surrounding rock structure was constructed to determine the final quantitative expression form.

[0067] Specifically, for the sparsity analysis and quantitative expression of the fracture network, the surrounding rock fracture distribution image is first preprocessed using digital image processing technology. Assuming that the resolution of the acquired surrounding rock image is 2000x2000 pixels, grayscale processing and edge detection algorithms (such as the Canny algorithm, with thresholds set to 100 and 200) are used to extract the fracture edges, generating a binary image in which the fracture pixel value is 1 and the background is 0. Next, based on the sparsity quantitative analysis method, the distribution density of the fracture pixels is calculated. Assuming that the proportion of fracture pixels in the total pixel points is 0.05, the spatial grid is divided (grid size is 50x50 pixels) and the proportion of fracture pixels in each grid is counted to obtain the sparsity distribution matrix. It is found that the fracture density of the grid in the central area is 0.12, while that in the edge area is only 0.03, indicating that the distribution is non-uniform. Subsequently, a sparsity eigenvector was constructed, and the principal component analysis (PCA) algorithm was used to reduce the dimensionality of the sparsity distribution matrix. The first three principal components were extracted, and their contribution rates were assumed to be 60%, 25%, and 10%, respectively. This formed the eigenvector [0.6, 0.25, 0.1], which was used to characterize the sparsity characteristics of the fracture network. Finally, the quantitative expression of the fracture network in the heterogeneity of the surrounding rock structure was determined. By calculating the correlation between the eigenvector and the mechanical parameters of the surrounding rock (such as the elastic modulus, assumed to be in the range of 20-50 GPa), the Pearson correlation coefficient algorithm was used to obtain a correlation coefficient of 0.85, indicating that fracture sparsity is highly correlated with the heterogeneity of the surrounding rock. The eigenvector was then combined with the mechanical parameters to form a quantitative model, providing data support for subsequent stability analysis.

[0068] In step S103, based on the association between the sparsity characteristic vector and the heterogeneity of the surrounding rock structure, the stratigraphic lithology and tectonic stress information in the historical geological data are obtained, and the stress changes and hydrogeological parameters in the real-time monitoring information are combined to obtain the preliminary integration results of the multi-source data.

[0069] Historical geological data is used to obtain stratum lithology and tectonic stress information. Through data cleaning and format standardization, a structured geological information dataset is generated. Time series analysis methods are used to analyze stress changes and hydrogeological parameters in real-time monitoring data to extract the dynamic characteristics of stress changes and the fluctuation patterns of hydrogeology, thereby identifying key trends in real-time data. Based on these key trends in real-time data and combined with the structured geological information dataset, a mapping relationship between sparsity feature vectors and surrounding rock structural heterogeneity is constructed to determine potential correlations between the two. If the correlation strength in the mapping relationship falls below a preset threshold, multi-source data fusion technology is used to perform a weighted integration of the historical geological data and real-time monitoring data to generate a comprehensive feature dataset.

[0070] Specifically, in order to study the correlation between sparsity eigenvectors and the heterogeneity of surrounding rock structure, we first extracted stratigraphic lithology and tectonic stress information from historical geological databases through data mining technology. For example, the stratigraphic lithology distribution was obtained from historical data of a mining area, showing that the overlying rock layer was mainly sandstone with a compressive strength of 50.2 MPa, while the underlying rock layer was mudstone with a compressive strength of only 20.5 MPa. At the same time, tectonic stress data was extracted, showing that the principal stress direction was northeast and the maximum principal stress value was 15.3 MPa. The principal component analysis algorithm was used to reduce the dimensionality of these data, extract eigenvectors, and obtain a sparse characteristic matrix, in which non-zero elements accounted for only 12.3%, reflecting the heterogeneous characteristics of the surrounding rock structure. Next, real-time data was acquired through a sensor network, combining stress changes and hydrogeological parameters from real-time monitoring information. For example, at a monitoring point, the stress change over the past 24 hours was 2.1 MPa, the groundwater level rose by 0.5 m, and the permeability coefficient was 1.2e-5 m / s. A time series analysis algorithm was used to predict the trend of these data, and the correlation coefficient between the stress change rate and the water level rise was calculated to be 0.78, indicating that hydrogeological conditions have a significant impact on surrounding rock stress. Finally, a multi-source data integration was performed on historical and real-time data using a weighted fusion algorithm, assigning a weight of 0.6 to historical data and 0.4 to real-time data. The resulting composite stress impact index was 9.8. A spatial interpolation method was then used to generate a map of the surrounding rock heterogeneity distribution, revealing that stress concentration areas overlap with lithologic weak zones at a rate of 85%, providing data support for subsequent stability analysis.

[0071] Step S104 , based on the preliminary integration results of the multi-source data, dynamic data integration technology is used to perform time series matching processing on the historical geological data and the real-time monitoring information to determine a unified feature set after the fusion of the multi-source data.

[0072] Based on the preliminary results of multi-source data integration, a time series analysis method was used to match historical geological data and real-time monitoring information to construct a unified feature set. Based on this unified feature set, interpolation methods were used to align the data based on the time series differences between the historical geological data and the real-time monitoring information, resulting in an aligned feature dataset. The distribution characteristics of the geological and monitoring information within this aligned feature dataset were determined, and the correlation strength between the data was determined. If the correlation strength fell below a preset threshold, the data was fused a second time to determine a fused comprehensive feature set. Based on this fused comprehensive feature set, a support vector machine algorithm was used to classify potential correlation patterns between the geological and monitoring information, resulting in a classified feature distribution.

[0073] Specifically, through the preliminary integration of multi-source data, historical geological data (such as stratum subsidence records) and real-time monitoring information (such as GPS displacement data) are first cleaned and formatted.

[0074] For example, historical geological data contains the annual settlement of a certain area from 2000 to 2020, in millimeters, and the data format is CSV, with some records missing. The K nearest neighbor algorithm (K=5) is used to interpolate the missing values, calculate the Euclidean distance between each missing point and the 5 adjacent complete records, take the weighted average to fill in, and obtain a complete data set. The real-time monitoring data is hourly GPS displacement in centimeters, with a time resolution of 1 hour, and the data format is JSON. Through timestamp alignment, the real-time data is aggregated according to the daily average and converted into millimeter units consistent with the historical data, with a unified time granularity of years. Next, the dynamic time warping (DTW) algorithm is used for time series matching to calculate the time offset and similarity between the two sets of data.

[0075] For example, a DTW distance of 150.23 indicates that the two series need to be aligned on the time axis by approximately three months to achieve optimal alignment. This alignment generates an aligned time series. Subsequently, a unified feature set was extracted, including sedimentation rate (mm / year), displacement variance, and periodicity. Principal component analysis (PCA) was used to reduce the feature dimensionality, retaining the top three principal components that contributed 95% of the cumulative variance.

[0076] For example, the original feature set contains 10 variables, and the feature vector after dimensionality reduction is [0.72, 0.19, 0.09], forming a fused feature set. Finally, the fused data is smoothed using a Kalman filter to reduce the influence of noise. The filter parameters Q = 0.01 and R = 0.1, resulting in a smoothed unified feature set suitable for subsequent geological disaster prediction models.

[0077] In step S105, a feature mapping relationship of geomechanical conditions is constructed based on the unified feature set after fusion of multi-source data. In combination with the sparsity feature vector, it is determined whether the input parameters of the surrounding rock stability prediction model meet the preset threshold requirements. If so, the initial prediction model parameters are generated.

[0078] In view of the multi-source data and fusion characteristics, data integration tools are used to standardize geological information from different sources to obtain a standardized feature data set. Based on the standardized feature data set, geomechanics and feature mapping are combined to construct a feature correlation matrix of mechanical conditions to determine the dependency relationships between features. In view of the dependency relationships between features, sparse representation methods are used to reduce the dimension of the feature matrix, combining sparse vectors and surrounding rock stability, to obtain a streamlined feature vector set. Based on the streamlined feature vector set, the prediction model and input parameters are combined, and the feature vectors are matched using parameter screening tools to determine whether they meet the preset threshold requirements. If so, a preliminary model parameter set is generated. Based on the preliminary model parameter set, the support vector machine algorithm is used to optimize and adjust the parameter set in combination with the threshold requirements and initial parameters to obtain the adjusted model parameter combination.

[0079] Specifically, based on a unified feature set derived from multi-source data fusion, assuming the initial data comes from geological exploration, seismic velocity testing, and borehole sampling, including rock strength (50 MPa), fracture density (0.2 fractures / m), groundwater seepage rate (0.1 L / min), and stress field distribution (principal stress 20 MPa), a weighted average fusion algorithm was first used, with weights of 0.4, 0.3, 0.2, and 0.1, to calculate the unified eigenvalue: 0.4 × 50 + 0.3 × 0.2 + 0.2 × 0.1 + 0.1 × 20 = 22.06. A feature mapping relationship for the geomechanical conditions was then constructed, and principal component analysis (PCA) was used to extract the first two principal components, which contributed 85%. The feature set was then reduced to a two-dimensional vector [22.06, 15.32]. To introduce sparsity, L1 regularization (Lasso regression, penalty coefficient λ = 0.5) is used to screen out key features and obtain a sparse vector [22.06, 0], indicating that the crack density has little effect on stability. To determine whether the input parameters meet the threshold requirements, the preset threshold is the principal component vector modulus > 20, and the modulus is calculated as √(22.06 2 +15.32 2) = 26.82 > 20, meeting the requirements. Initial prediction model parameters were generated using a support vector machine (SVM) classification model with an RBF kernel function, parameters C = 1.0, γ = 0.1, an initial hyperplane normal vector of [0.85, 0.52], and a bias b = 0.3. The analysis process demonstrated that data fusion ensured feature consistency, PCA dimensionality reduction preserved key information, Lasso screening enhanced model interpretability, and SVM parameter initialization provided a reliable foundation for subsequent surrounding rock stability prediction.

[0080] Step S106, based on the initial prediction model parameters, obtain the geomechanical condition related data of the unmined area, simulate and analyze the surrounding rock stability through intelligent prediction technology, and obtain the stability prediction distribution of the unmined area.

[0081] Basic data is obtained from the geomechanical conditions of unmined areas. Data integration tools are used to standardize the collected conditional data to obtain a standardized geological dataset. Based on the standardized geological dataset, the initial parameters and prediction model are combined, and feature extraction tools are used to preliminarily screen the data to determine the key feature set related to surrounding rock stability. Based on this key feature set, combined with intelligent prediction technology, a support vector machine algorithm is used to classify the feature set to obtain preliminary classification results for surrounding rock stability in unmined areas. Based on this preliminary classification result, combined with stability distribution and regional division, the classification results are regionalized using spatial mapping tools to determine the stability distribution characteristics of different regions. Based on the stability distribution characteristics, combined with the prediction results and geomechanics, if the distribution characteristics do not match the preset threshold, the feature data is adjusted using data correction tools to obtain the corrected distribution characteristics.

[0082] Specifically, to obtain the initial prediction model parameters, the system first extracts geological data from unmined areas through the geological exploration database. Assuming that the average compressive strength of rocks in unmined areas of a mining area is 50.2 MPa, the Poisson's ratio is 0.25, and the internal friction angle is 30 degrees, these parameters are automatically imported into the system through historical exploration data. A data cleaning algorithm is used to remove outliers (such as compressive strength values ​​that exceed the reasonable range by 10%), and the weighted average method is used to calculate the representative parameters of the region to form the initial model input. Subsequently, relevant data on the geomechanical conditions of the unmined area are obtained. The system uses a sensor network to collect ground stress data in real time. Assuming that the maximum principal stress in a certain area is 12.5 MPa and the minimum principal stress is 5.3 MPa, a three-dimensional geological model is constructed using finite element analysis software (such as ANSYS). The collected data is combined to perform grid division, and the grid unit size is set to 1 meter × 1 meter × 1 meter to ensure calculation accuracy.

[0083] Step S107: Based on the stability prediction distribution of the unmined area and the results of dynamic data integration, if abnormal stress changes are detected in the real-time monitoring information, the prediction distribution is locally corrected to determine the final stability prediction result.

[0084] Real-time monitoring data is acquired from monitoring equipment, and time series analysis methods are used to determine whether abnormal stress changes exist, thereby obtaining stress anomaly data. If stress anomaly data is detected, the degree of anomaly is determined using a preset threshold, and the K-means clustering algorithm is used to classify the anomaly data and determine the scope of the abnormal area. Based on the scope of the abnormal area, dynamic data of the unmined area is acquired from the database, and data fusion technology is used to integrate the dynamic data and real-time monitoring data to obtain a comprehensive data set. Based on the comprehensive data set, the support vector machine algorithm is used to analyze the stability prediction distribution of the unmined area to obtain the initial stability prediction distribution. If an abnormal area exists in the initial stability prediction distribution, the abnormal area is locally corrected, and the prediction distribution parameters are adjusted using the weighted average method to obtain the corrected prediction distribution. Using the corrected prediction distribution, the probability density estimation method is used to determine the final stability prediction result for the unmined area.

[0085] Specifically, in predicting the stability distribution of unmined areas, an initial prediction model is first constructed using geological exploration data and historical mining records. Assuming the average rock strength in an unmined area of ​​a mining area is 50 MPa, the stress distribution of each grid cell is calculated using finite element analysis methods, resulting in a stability index ranging from 0.6 to 0.9. Areas below 0.7 are marked as potentially unstable, accounting for approximately 15%. Subsequently, when integrating dynamic data, displacement and stress data collected in real time by sensors are used. Assuming that the displacement rate in a certain area suddenly increases from 0.1 mm / day to 0.5 mm / day, a time series analysis algorithm, such as the ARIMA model, is used to predict the displacement trend over the next 24 hours. The results are then compared with the initial model, and the calculated deviation value reaches 20%, triggering an abnormality warning. Next, when real-time monitoring information detects an abnormal stress change, such as a sudden increase in stress at a certain point from 30 MPa to 45 MPa, exceeding the safety threshold of 40 MPa, the system automatically initiates a local correction mechanism. Using a Bayesian update algorithm, it lowers the stability index within a 5-meter radius of the anomaly by 10%. It also recalculates the stress distribution of the affected grid, resulting in a revised stability index as low as 0.55. Finally, the correction results are combined with the global model to generate a final stability prediction distribution map, showing an increase in the proportion of unstable areas from 15% to 18%. Data visualization technology also marks high-risk areas in red to assist in subsequent mining planning.

[0086] Step S108: Based on the final stability prediction result, a geomechanical condition risk distribution map of the unmined area is generated through intelligent prediction technology to obtain the distribution characteristics of the potential risk area and complete the output of the prediction model.

[0087] An initial dataset is constructed by extracting geomechanical parameters from existing data. This data is then classified using a support vector machine algorithm to obtain preliminary geomechanical condition classification results. Based on these classification results, a risk assessment matrix is ​​constructed for the geomechanical conditions in unmined areas to obtain preliminary distribution characteristics of potential risks and determine an initial risk level distribution map. Spatial interpolation of this initial map is performed to generate a risk distribution map for the unmined areas, providing more detailed regional distribution characteristics and determining the boundaries of potential risk areas. If the risk level of certain areas in the risk distribution map exceeds a preset threshold, in-depth analysis of the geomechanical data in these areas is conducted to determine the specific locations of high-risk areas. Based on the distribution of high-risk areas and combined with stability prediction data, targeted risk identification results are generated to determine the demarcation of key monitoring areas. By integrating the demarcation of key monitoring areas with the output of the prediction model, final risk distribution data is generated, providing complete information on the distribution of potential risk areas. If the final distribution contains missing data or outliers, spatial neighborhood analysis is used to supplement the missing data to obtain a complete risk distribution map.

[0088] Specifically, based on the final stability prediction results, intelligent prediction technology is used to generate a geomechanical risk distribution map for unmined areas and identify the distribution characteristics of potential risk areas. This is achieved through the following technical process. First, geomechanical data, including rock strength (compressive strength 50-200 MPa), geostress field (principal stress 10-30 MPa), and fracture density (0.1-2 fractures / meter), is collected through geological exploration and sensor acquisition and stored in a database. Next, a random forest algorithm is used to predict stability. 100 decision trees are selected with a maximum depth of 20. Feature importance is assessed based on the Gini index. A training set to test set ratio of 8:2 is used. The model achieves an accuracy of over 85% and outputs a stability probability distribution (0-1). Subsequently, a convolutional neural network (CNN) is used to generate the risk distribution map. The input is a 256×256 pixel grid of geological parameters, a convolution kernel size of 3×3, a ReLU activation function, and training for 100 epochs. The loss function is the mean squared error. A risk probability heat map (risk value 0-100%) is generated. The heat map was then partitioned using the K-means clustering algorithm (K=3) to identify high-risk (>70%), medium-risk (30-70%), and low-risk (<30%) areas. Cluster centers were calculated using Euclidean distance, and convergence was achieved after 10 iterations. Finally, a report was generated containing the distribution characteristics of the risk areas, with high-risk areas accounting for 15%, medium-risk areas accounting for 45%, and low-risk areas accounting for 40%. Verification was also conducted using geological fault data (fault density 0.05 per square kilometer), confirming that high-risk areas overlap with fault-dense areas by 80%.

[0089] On the other hand, this embodiment further provides an electronic device, comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the method is implemented when the processor executes the computing program.

[0090] On the other hand, this embodiment further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the method when executed by a processor.

[0091] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An intelligent analysis method for the geomechanical conditions of coal seam surrounding rocks, characterized in that: include: Digitally process image data and geological exploration records of the surrounding rock fracture network to obtain a preliminary characteristic dataset of fracture distribution; Based on the preliminary feature data set, a sparsity quantitative analysis method is used to deeply mine the crack distribution characteristics and construct a sparsity feature vector; By combining the correlation between the sparsity characteristic vector and the heterogeneity of the surrounding rock structure, the stratigraphic lithology and tectonic stress information in the historical geological data, as well as the stress changes and hydrogeological parameters in the real-time monitoring information, are obtained to obtain the preliminary integration results of the multi-source data; Performing time series matching processing on the historical geological data and real-time monitoring information in the preliminary integration results by using dynamic data integration technology to determine a unified feature set after multi-source data fusion; Based on the unified feature set, a feature mapping relationship of geomechanical conditions is constructed, and combined with the sparsity feature vector, whether the input parameters of the surrounding rock stability prediction model meet the preset threshold requirements is determined, and if so, initial prediction model parameters are generated; Obtain relevant data on geomechanical conditions in unmined areas, simulate and analyze surrounding rock stability using intelligent prediction technology, and obtain predicted stability distribution in unmined areas; If abnormal stress changes are detected in the real-time monitoring information, the stability prediction distribution is locally corrected to determine the final stability prediction result; Based on the final stability prediction results, a geomechanical condition risk distribution map of the unmined area is generated, the distribution characteristics of the potential risk area are obtained, and the output of the prediction model is completed.

2. The method according to claim 1, characterized in that The step of digitally processing the image data of the surrounding rock fracture network and the geological exploration records to obtain a preliminary characteristic data set of fracture distribution includes: Process network images and geological exploration data of surrounding rock fractures to obtain digitally converted raw data sets; Image processing technology is used to perform denoising and edge detection on the network image, extract preliminary contour information of crack distribution, and obtain an initial feature set of crack distribution.

3. The method according to claim 1, characterized in that The step of using a sparsity quantitative analysis method to deeply mine the crack distribution characteristics based on the preliminary feature data set and constructing a sparsity feature vector includes: Obtaining initial distribution information from the preliminary feature data set, and using a data preprocessing method to clean and format the raw data of the fracture network to obtain a preliminary fracture distribution data set; Based on the preliminary fracture distribution data set, a sparse characteristic analysis method is used to extract sparsity-related parameters and determine a sparse characteristic index in the fracture network; According to the sparse characteristic index, a quantitative analysis technology is used to construct a feature vector, obtain the sparsity feature vector expression of the fracture network, and construct a sparsity feature vector.

4. The method according to claim 1, wherein The step of combining the association between the sparsity characteristic vector and the surrounding rock structural heterogeneity to obtain the stratum lithology and tectonic stress information in the historical geological data, as well as the stress changes and hydrogeological parameters in the real-time monitoring information, to obtain the preliminary integration result of the multi-source data includes: Obtain stratum lithology and tectonic stress information from historical geological data, and obtain a structured geological information dataset through data cleaning and format standardization; Based on the stress changes and hydrogeological parameters in the real-time monitoring information, a time series analysis method is used to extract the dynamic characteristics of stress changes and the fluctuation laws of hydrogeology, and determine the key change trends of the real-time data; Based on the key change trends of the real-time data and in combination with the structured geological information dataset, a mapping relationship between the sparsity feature vector and the surrounding rock structural heterogeneity is constructed to determine the potential correlation between the two; If the correlation strength in the mapping relationship is lower than a preset threshold, the historical geological data and the real-time monitoring data are weightedly integrated through multi-source data fusion technology to obtain a preliminary integration of the multi-source data.

5. The method according to claim 1, wherein The step of performing time series matching processing on the historical geological data and the real-time monitoring information in the preliminary integration result by using the dynamic data integration technology to determine a unified feature set after the multi-source data is fused includes: Using a time series analysis method to match the historical geological data and real-time monitoring information in the preliminary integration results to construct a unified feature set; For the unified feature set, an interpolation method is used to align the data to obtain an aligned feature data set; For the aligned feature data set, the distribution characteristics of geological information and monitoring information are obtained, and the correlation strength between the data is determined. If the correlation strength is lower than a preset threshold, the data is subjected to secondary fusion processing to determine a unified feature set after multi-source data fusion.

6. The method according to claim 1, characterized in that The process of generating initial prediction model parameters includes: Using a data integration tool to standardize geological information from different sources in the unified feature set to obtain a standardized feature data set; Based on the normalized feature data set, combining geomechanics with feature mapping, constructing a feature correlation matrix of mechanical conditions to determine dependency relationships between features; Based on the dependency relationship between the features and the sparse feature vectors, a sparse representation method is applied to reduce the dimension of the feature matrix to obtain a simplified feature vector set; According to the simplified feature vector set, the feature vector is matched by a parameter screening tool to determine whether it meets the preset threshold requirements. If so, a preliminary model parameter set is generated; A support vector machine algorithm is used to optimize and adjust the preliminary model parameter set to obtain an adjusted model parameter combination; A verification process is performed based on the adjusted model parameter combination. If the verification result meets the preset stability standard, the final prediction model parameters are determined.

7. The method according to claim 1, characterized in that The process of determining the final stability prediction result includes: Obtain real-time monitoring data from monitoring equipment, use time series analysis methods to determine whether there are abnormal stress changes, and obtain stress anomaly data; If the stress abnormal data is detected, the abnormality degree is judged by a preset threshold, and the K-means clustering algorithm is used to classify the abnormal data to determine the range of the abnormal area; According to the range of the abnormal area, dynamic data of the unmined area is obtained from the database, and the dynamic data and real-time monitoring data are integrated using data fusion technology to obtain a comprehensive data set; For the comprehensive data set, a support vector machine algorithm is used to analyze the stability prediction distribution of the unmined area to obtain an initial stability prediction distribution; If there is an abnormal area in the initial stability prediction distribution, the abnormal area is locally corrected, and the prediction distribution parameters are adjusted using a weighted average method to obtain a corrected prediction distribution; The final stability prediction result of the unmined area is determined by using the modified prediction distribution and the probability density estimation method.

8. The method according to claim 1, characterized in that The steps of completing the output of the prediction model include: Extract geomechanical parameters from existing data, construct an initial data set, and use the support vector machine algorithm to classify the data to obtain preliminary geomechanical condition classification results. Based on the classification results, a risk assessment matrix is ​​constructed for the geomechanical conditions of the unmined area to obtain preliminary distribution characteristics of potential risks and determine an initial mapping of risk level distribution; By performing spatial interpolation processing on the initial map, a risk distribution map of the unmined area is generated to obtain more detailed regional distribution characteristics and determine the boundary range of the potential risk area; If the risk level of certain areas in the risk distribution map exceeds a preset threshold, the geomechanical data of these areas are deeply analyzed to obtain the specific distribution locations of high-risk areas; Based on the distribution of the high-risk areas and combined with stability prediction data, targeted risk identification results are generated to determine the scope of key monitoring areas; By integrating the scope division of the key monitoring areas with the output of the prediction model, the final risk distribution characteristic data is generated to obtain complete potential risk area distribution information; If there are missing data or abnormal points in the final distribution information, the missing parts are supplemented by the spatial neighborhood analysis method to obtain complete risk distribution map data.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein: When the processor executes the computing program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images