A three-dimensional visualization reconstruction method for overlying rock mass structure of deep coal mine working face

Abstract

The present invention discloses a method for three-dimensional visualization reconstruction of the overburden structure of a deep coal mine working face, including: 1) arranging a borehole of a mine borehole imager on the working face to collect and store basic geological data of the overburden layer; 2) the mine borehole imager outputs a borehole image; 3) extracting key geological features of the borehole image based on a constructed convolutional neural network for classification and identification to obtain information on rock layers, cracks, and faults; 4) horizontally linking vertical virtual boreholes along the deep coal mine working face and dividing the strata, using an improved Kriging spatial interpolation method, and building a three-dimensional geological model based on three-dimensional visualization software; 5) introducing gas concentration and stress distribution data of the monitoring area of ​​the deep coal mine working face, and realizing three-dimensional visualization of the overburden structure and risk area of ​​the deep coal mine working face through multimodal data fusion. The method of the present invention can realize accurate visualization of the overburden structure of the deep coal mine working face, and provide comprehensive geological data support for underground mine operations.

Description

Technical Field

[0001] The invention relates to the technical field of coal mine geological structure detection and three-dimensional visualization, and in particular to a three-dimensional visualization reconstruction method for overlying rock mass structure on a deep coal mine working face. Background Art

[0002] In coal mining engineering, the structural complexity and uncertainty of the roof of deep coal mine working faces pose significant challenges to the safety and efficiency of mining. Traditional coal mine geological exploration methods mainly rely on two-dimensional geological exploration data and empirical models, lacking high-resolution and high-precision spatial information. These methods have disadvantages such as low resolution, limited coverage, and inability to update in real time during data collection and analysis, which makes it impossible to accurately describe the structural characteristics of rock formations under complex geological conditions, thus affecting the safe mining of mines.

[0003] Especially in the three-dimensional geological structure modeling of coal mine roof, current research is still based on limited drilling data and simple interpolation methods, which cannot fully reflect important information such as lithology changes, fracture distribution, and structural stress characteristics. The shortcomings of existing technologies in data collection methods and model construction efficiency are mainly manifested in the following points:

[0004] 1. Insufficient data acquisition accuracy: Traditional data acquisition methods make it difficult to fully and accurately obtain the spatial structure information of the roof rock layer, resulting in low quality of basic data for model construction.

[0005] 2. Low efficiency of model construction: Existing 3D geological modeling methods mostly use manual drawing and empirical speculation, which is time-consuming and easily affected by subjective factors, and it is difficult to quickly adapt to the needs of complex geological conditions.

[0006] 3. Poor adaptability to complex structural areas: In complex structural areas such as faults and fracture zones, existing modeling methods are difficult to accurately capture subtle changes between rock layers, resulting in limited practical guidance significance of the model.

[0007] These deficiencies lead to significant safety hazards in mining design and roof management, such as tunnel collapse, rock burst, and gas explosion. Summary of the invention

[0008] The purpose of the present invention is to provide a three-dimensional visualization reconstruction method for the overlying rock mass structure of a deep coal mine working face in view of the deficiencies in the prior art. By combining a mine borehole imager and a convolutional neural network system, the rock stratum characteristics of the coal mine working face roof can be accurately identified and extracted, and the three-dimensional geological model can be reconstructed using the improved Kriging spatial interpolation technology. This can achieve accurate visualization of the overlying rock mass structure of the deep coal mine working face, provide comprehensive geological data support for underground operations, help improve the safety and efficiency of coal mining, reduce the occurrence of roof accidents and gas accidents, and provide scientific guidance for optimizing drilling construction plans and ensuring safe mining.

[0009] In order to achieve the above object, the present invention adopts the following technical solution.

[0010] A three-dimensional visualization reconstruction method for overlying rock mass structure of a deep coal mine working face comprises the following steps:

[0011] Step S1, arranging a borehole for a mine borehole imager on site at a deep coal mine working face, collecting basic geological data of the overburden layer, and storing the collected data;

[0012] Step S2, the mining borehole imager outputs a borehole wall distribution diagram and a three-dimensional column diagram according to the rock formation data recorded by the drilling;

[0013] Step S3, constructing a convolutional neural network, automatically extracting key geological features from the borehole wall distribution map and the three-dimensional column map output by the mine borehole imager based on the constructed convolutional neural network, and classifying and identifying the extracted key geological features to obtain rock layer, fracture, and fault information;

[0014] Step S4, based on the basic geological data of the overburden layer collected by drilling in step S1 and the key geological feature information obtained in step S3, vertical virtual boreholes are horizontally connected along the deep coal mine working face and the strata are divided, and a three-dimensional geological model is constructed by using an improved Kriging spatial interpolation method and relying on three-dimensional visualization software;

[0015] Step S5, introduce the gas concentration and stress distribution data of the deep coal mine working face monitoring area, spatially align the key geological feature information extracted by the convolutional neural network with the gas concentration and stress distribution data, and combine with the three-dimensional geological model constructed in step S4 to achieve three-dimensional visualization of the overlying rock structure and risk areas of the deep coal mine working face through multimodal data fusion.

[0016] Specifically, the drilling spacing of the drilling holes in step S1 is 200m, the drilling inclination angle is 45°, the drilling depth is between 42 and 45m, and the arrangement of the drilling holes covers the entire deep coal mine working face.

[0017] Specifically, the convolutional neural network in step S3 includes an input layer, a convolutional layer, a modified activation function, a pooling layer, a fully connected layer and an output layer;

[0018] The input layer is used to receive the original drilling image data and convert it into a numerical matrix form as a basic input for subsequent feature extraction;

[0019] The convolution layer slides on the input data through the convolution kernel and performs convolution operation to extract different levels of features of the image, including shallow convolution layer and deep convolution layer. The shallow convolution layer extracts basic features including edges and textures, and the deep convolution layer gradually abstracts complex features including rock layer shape and crack direction.

[0020] The improved activation function is improved on the basis of the traditional ReLU activation function, and the improved activation function expression is:

[0021] ;

[0022] In the above formula, It is a parameter between 0 and 1, and its value is adjusted according to the needs of geological feature identification; x It is the original feature value of each pixel in the feature map after the convolution operation. In the neural network, the role of the activation function is to perform nonlinear transformation on the feature map output by the convolution layer.

[0023] The pooling layer uses maximum pooling or average pooling operations to perform dimensionality reduction processing on the feature map after convolution, thereby reducing the computational complexity, retaining the main geological feature information, and enhancing the robustness of the convolutional neural network to image scale changes;

[0024] The fully connected layer is used to fully connect the pooled feature vector with the output of the convolutional neural network to map it to the target category corresponding to the geological feature;

[0025] The output layer is used to output the final geological feature classification and recognition results, including category labels such as rock layers, cracks, and faults and their corresponding location information.

[0026] Furthermore, the specific process of automatically extracting key geological features by the convolutional neural network in step S3 is as follows:

[0027] Step S31, convolution operation;

[0028] First, the parameters of the convolution kernel are initialized. The size, step length and number of the convolution kernel are determined according to the complexity of the geological features of the field exploration and the image resolution of the mining borehole imager. The original borehole image data converted into a numerical matrix is ​​input into the convolution kernel for convolution operation. Each pixel point of the image and the pixel value in its field are multiplied and summed with the corresponding weight of the convolution kernel to obtain the feature map after convolution. In the shallow convolution layer, basic features including the edge contour of the rock layer and the initial shape of the crack are extracted; the deep convolution layer extracts complex features including the layered structure of the rock layer, the extension direction and density of the crack.

[0029] Step S32, applying a modified activation function to the convolution result;

[0030] After each convolution operation such as step S31, a modified activation function is applied to perform nonlinear transformation on the convolution result to enhance the learning and expression ability of the convolutional neural network, so that the convolutional neural network can more accurately extract the complex distribution law of geological features;

[0031] Step S33: pooling operation and full connection;

[0032] After the nonlinear transformation, a pooling operation is performed. The maximum pooling or average pooling method is used to reduce the dimension of the convolution feature map according to a certain pooling window size and step size, thereby reducing the resolution of the feature map while retaining the main information of the geological features. The pooled feature vector is flattened and input into the fully connected layer, and the neurons of the fully connected layer are fully connected to each element of the flattened feature vector.

[0033] Step S34, classify and output;

[0034] According to the probability distribution of the output of the fully connected layer, the category with the largest probability is selected as the final geological feature classification result, and the classification result and the location information parameters of the corresponding geological features are output.

[0035] Specifically, in step S4, the improved Kriging spatial interpolation method is used to construct a three-dimensional geological model based on three-dimensional visualization software. The process is as follows:

[0036] The traditional Kriging spatial interpolation method is used to calculate the weight coefficient The solution equations are transformed into nonlinear functions, and the weight coefficients are obtained by solving the nonlinear function :

[0037] ;

[0038] In the above formula, An unknown point in space and The variance function value between ; For sampling point and The variance function value between i =1,2,3,...,n, j =1,2,3,...,n, where n is the number of known sample points that can affect the sampling point; is the Lagrange multiplier;

[0039] The calculation formula of the variance function value is:

[0040] ;

[0041] In the above formula, is the variation function value; h is the spatial distance; and The location and location The variable value at; E is the symbol for variance calculation;

[0042] The weight coefficient obtained Through equivalent infinitesimal transformation and Taylor formula expansion, we can get a high-order nonlinear function:

[0043] ;

[0044] The weight coefficient is obtained by differentiating the high-order nonlinear function The extreme value point, based on which the optimized weight coefficient is obtained , the optimized weight coefficient Substitute it into the basic equations of Kriging spatial interpolation method and deduce the weight coefficient The optimal value Z(x) of the unknown point is:

[0045] ;

[0046] The improved Kriging spatial interpolation method is used to perform linear unbiased optimal estimation of unknown points in the monitoring area through known sample points. Then, the discrete drilling data are interpolated layer by layer according to the divided strata, and the top and bottom surfaces of the strata are generated by three-dimensional visualization software. The surfaces are further combined into a body to construct a nonlinear three-dimensional geological model of the overlying rock mass structure of the deep coal mine working face.

[0047] Specifically, in step S5, three-dimensional visualization of the overlying rock mass structure and risk areas of the deep coal mine working face is achieved through multimodal data fusion, and the process is as follows:

[0048] Step S51, data preprocessing;

[0049] The convolutional neural network is used to extract geological feature information including rock layers, fractures and faults from the exploration borehole images, and record their parameters including position, size and shape. At the same time, the introduced gas concentration and stress distribution data are normalized and converted into a numerical range suitable for the exploration borehole image data.

[0050] Step S52: parameter quantization;

[0051] The number of cracks per unit length or unit area in the exploration borehole image is counted and recorded as For gas concentration, select the maximum gas concentration value in the monitoring area As a quantitative parameter; and determine the maximum stress value of the stress concentration point based on the stress distribution data And the area ratio of stress concentration area ;

[0052] Step S53: construct a comprehensive risk assessment algorithm model;

[0053] The expression of the comprehensive risk assessment algorithm model is:

[0054] ;

[0055] In the above formula, represents the comprehensive risk value, , , They are the weight coefficients of the number of fractures, the maximum gas concentration and the maximum stress, which are determined according to the actual geological conditions and mining experience; the adjustment coefficient K=0.5;

[0056] Considering the impact of the area ratio of stress concentration areas on risk, the comprehensive risk assessment algorithm model is modified to obtain the comprehensive risk value :

[0057] ;

[0058] In the above formula, k It is an adjustment coefficient used to regulate the influence of the proportion of stress concentration area on risk.

[0059] Specifically, the comprehensive risk value ≥0.8, the monitored area is a high-risk area; the comprehensive risk value is 0.5≤ <0.8, the monitored area is a medium-risk area; the comprehensive risk value <0.5, the monitored area is a low-risk area.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The method of the present invention combines a mining borehole imager and a convolutional neural network system to accurately identify and extract the rock stratum characteristics of the overlying rock structure on the deep coal mine working face, and reconstructs the three-dimensional geological model using the improved Kriging spatial interpolation technology. It can realize the accurate visualization of the overlying rock structure on the deep coal mine working face, provide comprehensive geological data support for underground drilling operations, help improve the safety and efficiency of coal mining, reduce the occurrence of roof accidents and gas accidents, and provide scientific guidance for optimizing drilling construction plans and ensuring safe mining. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a flow chart of a method for three-dimensional visualization reconstruction of overlying rock mass structure on a deep coal mine working face of the present invention;

[0063] Figure 2 is a schematic diagram of virtual drilling holes being connected laterally along a working surface in an embodiment of the present invention;

[0064] Figure 3 is a schematic diagram of stratum division in an embodiment of the present invention;

[0065] Figure 4 is a schematic diagram of the stratigraphic division of the pinch-out layer in an embodiment of the present invention;

[0066] Figure 5 is a schematic diagram of interpolating a surface in an embodiment of the present invention;

[0067] Figure 6 is a schematic diagram of a body enclosed by the surfaces in an embodiment of the present invention;

[0068] Figure 7 It is a schematic diagram of a three-dimensional geological model of a deep coal mine working face constructed in an embodiment of the present invention.

[0069] In the figure: 101, working face; 102, test drilling hole; 103, virtual drilling hole; 104, return air lane; 105, machine lane; 106, stratigraphic boundary; 107, lens; 108, pinch-out. DETAILED DESCRIPTION

[0070] In order to facilitate those of ordinary skill in the art to understand and implement the present invention, the following detailed description is given to each step of the method proposed by the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

[0071] Example

[0072] like Figure 1As shown, the present invention discloses a three-dimensional visualization reconstruction method of the overlying rock mass structure on a deep coal mine working face, comprising the following steps:

[0073] Step S1, arranging a borehole for a mine borehole imager on site at a deep coal mine working face, collecting basic geological data of the overburden layer, and storing the collected data;

[0074] In this embodiment, a CXK12 (B) mining borehole imaging device is used to arrange the drilling camera work of the deep coal mine working face 101. A total of 11 boreholes are designed, and one borehole is arranged every 200m along the entire deep coal mine return air lane 104 or machine lane 105. The borehole is inclined at 45° to the roof of the 2# coal seam, and the borehole depth is between 42 and 45m. It passes through the 1# coal seam, and the lithology, fractures, and thickness changes and occurrence of each rock layer in the borehole are observed throughout the process.

[0075] It should be noted that the CXK12(B) mining borehole imager collects rock layer images through drilling, and records the geological information of the roof rock layer in real time, obtains high-resolution image data, and describes in detail the geological characteristics such as rock layers, cracks, faults, etc.; during the drilling imaging process, all collected image data are stored in real time to the data storage unit through the operating system of the CXK12(B) mining borehole imager. At the same time, the built-in data processing module of the CXK12(B) mining borehole imager will process the initially collected images, automatically adjust parameters to optimize image quality, and make adaptive adjustments according to the working face conditions to ensure that the acquired image data is accurate and clear, which is convenient for subsequent analysis.

[0076] Step S2, the mining borehole imager outputs a borehole wall distribution diagram and a three-dimensional column diagram according to the rock formation data recorded by the drilling;

[0077] During the borehole imaging process, the image data collected by the CXK12(B) mining borehole imager is processed by the supporting software. First, noise removal, enhancement and contrast adjustment are performed to ensure the clarity and accuracy of the image. Then, the software generates borehole wall distribution maps based on the processed images. These maps show the thickness of the rock layer, the direction of the cracks and the distribution of the faults. In addition, the software also generates a three-dimensional bar graph to intuitively represent the spatial distribution and structural characteristics of the rock layer at different depths.

[0078] Step S3, constructing a convolutional neural network, automatically extracting key geological features from the borehole wall distribution map and the three-dimensional column map output by the mine borehole imager based on the constructed convolutional neural network, and classifying and identifying the extracted key geological features to obtain rock layer, fracture, and fault information;

[0079] The convolutional neural network includes an input layer, a convolution layer, an improved activation function, a pooling layer, a fully connected layer and an output layer, wherein:

[0080] The input layer is used to receive the original drilling image data and convert it into a numerical matrix form as the basic input for subsequent feature extraction;

[0081] The convolution layer slides on the input data and performs convolution operations through the convolution kernel to extract different levels of image features, including shallow convolution layers and deep convolution layers. The shallow convolution layers extract basic features including edges and textures, and the deep convolution layers gradually abstract complex features including rock layer shapes and crack directions.

[0082] The improved activation function is improved on the basis of the traditional ReLU activation function. By giving a certain weight to the negative part, it can more accurately identify and distinguish geological features. The expression of the improved activation function is:

[0083] ;

[0084] In the above formula, It is a parameter between 0 and 1, and its value is adjusted according to the needs of geological feature identification; x It is the original feature value of each pixel in the feature map after the convolution operation. In the neural network, the role of the activation function is to perform nonlinear transformation on the feature map output by the convolution layer.

[0085] By adjusting the parameter α in the activation function, the degree of attention paid to negative features can be controlled. For example, when identifying geological features such as cracks that may appear as dark areas in the image, appropriately increasing the α value helps to highlight the contribution of these features in the negative part, thereby improving the accuracy of recognition.

[0086] The pooling layer uses maximum pooling or average pooling operations to reduce the dimension of the feature map after convolution, reduce the computational complexity, retain the main geological feature information, and enhance the robustness of the convolutional neural network to image scale changes;

[0087] The fully connected layer is used to fully connect the pooled feature vector with the output of the convolutional neural network to map it to the target category corresponding to the geological feature;

[0088] The output layer is used to output the final geological feature classification and identification results, including category labels such as rock layers, cracks, and faults and their corresponding location information.

[0089] Furthermore, the process of automatically extracting key geological features using the convolutional neural network is as follows:

[0090] Step S31, convolution operation;

[0091] First, the parameters of the convolution kernel are initialized. The size, step length and number of the convolution kernel are determined according to the complexity of the geological features of the field exploration and the image resolution of the mining borehole imager. The original borehole image data converted into a numerical matrix is ​​input into the convolution kernel for convolution operation. Each pixel point of the image and the pixel value in its field are multiplied and summed with the corresponding weight of the convolution kernel to obtain the feature map after convolution. In the shallow convolution layer, basic features including the edge contour of the rock layer and the initial shape of the crack are extracted; the deep convolution layer extracts complex features including the layered structure of the rock layer, the extension direction and density of the crack.

[0092] Step S32, applying a modified activation function to the convolution result;

[0093] After each convolution operation such as step S31, a modified activation function is applied to perform nonlinear transformation on the convolution result to enhance the learning and expression ability of the convolutional neural network, so that the convolutional neural network can more accurately extract the complex distribution law of geological features;

[0094] Step S33: pooling operation and full connection;

[0095] After the nonlinear transformation, a pooling operation is performed, and a maximum pooling or average pooling method is used to perform dimensionality reduction processing on the feature map after convolution according to a certain pooling window size and step size. For example, for a pooling window of size 3×3, a step size of 1 is used to slide on the feature map, and the maximum value or average value in the pooling window is taken as the output feature value of the corresponding position of the window, thereby reducing the resolution of the feature map and reducing the amount of calculation, while retaining the main information of the geological features, and improving the calculation efficiency and generalization ability of the model; the feature vector after pooling is flattened and input into the fully connected layer, and the neurons of the fully connected layer are fully connected to each element of the flattened feature vector. In the fully connected layer, the input feature vector is mapped to the target category space, that is, the classification label of the geological feature, through the learned weight matrix and bias vector. In this embodiment, the corresponding number of output neurons is designed for different geological feature categories, and the output result is subjected to probability distribution processing using the softmax function to determine the probability that each geological feature belongs to a certain category;

[0096] Step S34, classify and output;

[0097] According to the probability distribution of the output of the fully connected layer, the category with the highest probability is selected as the final geological feature classification result, and the classification result and the location information parameters of the corresponding geological features are output so as to accurately locate these features to the corresponding spatial positions in the subsequent three-dimensional reconstruction process.

[0098] Step S4, based on the basic geological data of the overburden layer collected by drilling in step S1 and the key geological feature information obtained in step S3, vertical virtual boreholes are horizontally connected along the deep coal mine working face and the strata are divided, and a three-dimensional geological model is constructed by using an improved Kriging spatial interpolation method and relying on three-dimensional visualization software;

[0099] Step S41: Draw vertical virtual boreholes horizontally along the deep coal mine working face and divide the strata

[0100] According to the drilling image data, such as Figure 2 As shown, combining geological theory and actual geological conditions, the test borehole 102 is horizontally connected along the working surface according to the lithology to obtain a vertical virtual borehole 103; in order to prevent the occurrence of stratigraphic layer intersection in subsequent spatial interpolation, as shown in FIG. Figure 3 As shown, the virtual boreholes are spread in three-dimensional space, and artificial comparison is performed according to the borehole stratigraphic information to divide the stratigraphic boundaries 106; adjacent boreholes are compared layer by layer to determine whether they belong to the same layer, especially the thickness change of the same stratigraphic layer between different boreholes. If a stratigraphic layer disappears or becomes thinner in a certain borehole, and the layer continues to exist in the adjacent borehole, it can be determined that the layer is a pinch-out stratigraphic layer, and the specific pinch-out position is obtained by interpolation; for special geological conditions such as lens 107 and pinch-out 108, such as Figure 4 As shown, in the stratigraphic division, the CD segment should be regarded as a pinch-out layer with a thickness of 0, that is, the ACD segment is the upper surface of the pinch-out layer, and the BCD segment is the lower surface of the pinch-out layer. Similarly, the division of all stratigraphic layers is completed;

[0101] Step S42, using the improved Kriging spatial interpolation method and relying on three-dimensional visualization software to construct a three-dimensional geological model;

[0102] The traditional Kriging spatial interpolation method is a spatial interpolation method based on statistics. The core idea is to perform a linear unbiased optimal estimation of unknown points through known sample points. However, the core theory of the traditional Kriging spatial interpolation method is the estimation of linear sample points. The variance of the estimated value is smaller than the variance of the original data. When the sample points are small and unevenly distributed, a large estimation error may occur. Therefore, in this embodiment, the traditional Kriging spatial interpolation method is improved so that it can perform the optimal evaluation of nonlinear sample points. The specific improvement method and process are as follows:

[0103] The traditional Kriging spatial interpolation method is used to calculate the weight coefficient The solution equations are transformed into nonlinear functions, and the weight coefficients are obtained by solving the nonlinear function :

[0104] ;

[0105] In the above formula, An unknown point in space and The variance function value between ; For sampling point and The variance function value between i =1,2,3,...,n, j =1,2,3,...,n, where n is the number of known sample points that can affect the sampling point; is the Lagrange multiplier;

[0106] The calculation formula of the variance function value is:

[0107] ;

[0108] In the above formula, is the variation function value; h is the spatial distance; and The location and location The variable value at; E is the symbol for variance calculation;

[0109] The weight coefficient obtained Through equivalent infinitesimal transformation and Taylor formula expansion, we can get a high-order nonlinear function:

[0110] ;

[0111] The weight coefficient is obtained by differentiating the high-order nonlinear function The extreme value point, based on which the optimized weight coefficient is obtained , the optimized weight coefficient Substitute it into the basic equations of Kriging spatial interpolation method and deduce the weight coefficient The optimal value Z(x) of the unknown point is:

[0112] ;

[0113] The improved Kriging spatial interpolation method is used to perform linear unbiased optimal estimation of unknown points in the monitoring area through known sample points, such as Figure 5 As shown in the figure, the top and bottom surfaces of the strata are generated by interpolating the discrete borehole data layer by layer based on the divided strata and relying on the EVS 3D geological information visualization software. Figure 6 As shown, the surfaces are enclosed into a body and the following is constructed: Figure 7 The nonlinear three-dimensional geological model of the overlying rock structure of the deep coal mine working face is shown.

[0114] Step S5, introducing the gas concentration and stress distribution data of the deep coal mine working face, spatially registering the key geological feature information extracted by the convolutional neural network with the gas concentration and stress distribution data, combining the three-dimensional geological model constructed in step S4, and realizing the three-dimensional visualization of the overlying rock structure and risk area of ​​the deep coal mine working face through multimodal data fusion;

[0115] Step S51, data preprocessing;

[0116] The convolutional neural network is used to extract geological feature information including rock layers, fractures and faults from the exploration borehole images, and record their parameters including position, size and shape. At the same time, the introduced gas concentration and stress distribution data are normalized and converted into a numerical range suitable for the exploration borehole image data.

[0117] Step S52: parameter quantization;

[0118] The number of cracks per unit length or unit area in the exploration borehole image is counted and recorded as For gas concentration, select the maximum gas concentration value in the monitoring area As a quantitative parameter; and determine the maximum stress value of the stress concentration point based on the stress distribution data And the area ratio of stress concentration area ;

[0119] Step S53: construct a comprehensive risk assessment algorithm model;

[0120] The expression of the comprehensive risk assessment algorithm model is:

[0121] ;

[0122] In the above formula, represents the comprehensive risk value, , , are the weight coefficients of the number of cracks, the maximum gas concentration and the maximum stress, respectively, which are determined according to the actual geological conditions and mining experience. In this embodiment, the weight coefficients are determined according to the actual situation of the deep coal mine working face. , , ; Adjustment coefficient K=0.5;

[0123] Considering the impact of the area ratio of stress concentration areas on risk, the comprehensive risk assessment algorithm model is modified to obtain the comprehensive risk value :

[0124] ;

[0125] In the above formula,k It is an adjustment coefficient used to regulate the influence of the proportion of stress concentration area on risk.

[0126] The comprehensive risk value calculated according to the comprehensive risk assessment model , mark and grade the risk areas on the visual interface, and display the comprehensive risk value ≥0.8, the monitored area is a high-risk area; the comprehensive risk value is 0.5≤ <0.8, the monitored area is a medium-risk area; the comprehensive risk value <0.5, the monitored area is shown as a low-risk area, thus providing guidance for roof management and safety warning during coal mining.

[0127] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A three-dimensional visualization reconstruction method for the overlying rock mass structure of a deep coal mine working face, characterized in that: The following steps are involved: Step S1, arranging a borehole for a mine borehole imager on site at a deep coal mine working face, collecting basic geological data of the overburden layer, and storing the collected data; Step S2, the mining borehole imager outputs a borehole wall distribution diagram and a three-dimensional column diagram according to the rock formation data recorded by the drilling; Step S3, constructing a convolutional neural network, automatically extracting key geological features from the borehole wall distribution map and the three-dimensional column map output by the mine borehole imager based on the constructed convolutional neural network, and classifying and identifying the extracted key geological features to obtain rock layer, fracture, and fault information; Step S4, based on the basic geological data of the overburden layer collected by drilling in step S1 and the key geological feature information obtained in step S3, vertical virtual boreholes are horizontally connected along the deep coal mine working face and the strata are divided, and a three-dimensional geological model is constructed by using an improved Kriging spatial interpolation method and relying on three-dimensional visualization software; Step S5, introduce the gas concentration and stress distribution data of the deep coal mine working face monitoring area, spatially align the key geological feature information extracted by the convolutional neural network with the gas concentration and stress distribution data, and combine with the three-dimensional geological model constructed in step S4 to achieve three-dimensional visualization of the overlying rock structure and risk areas of the deep coal mine working face through multimodal data fusion.

2. The method for three-dimensional visualization reconstruction of overlying rock mass structure of a deep coal mine working face according to claim 1, characterized in that: In step S1, the drilling spacing of the boreholes is 200m, the drilling inclination angle is 45°, the drilling depth is between 42 and 45m, and the arrangement of the boreholes covers the entire deep coal mine working face.

3. The method for three-dimensional visualization reconstruction of overlying rock mass structure on a deep coal mine working face according to claim 1, characterized in that: The convolutional neural network in step S3 includes an input layer, a convolutional layer, a modified activation function, a pooling layer, a fully connected layer and an output layer; The input layer is used to receive the original drilling image data and convert it into a numerical matrix form as a basic input for subsequent feature extraction; The convolution layer slides on the input data through the convolution kernel and performs convolution operation to extract different levels of features of the image, including shallow convolution layer and deep convolution layer. The shallow convolution layer extracts basic features including edges and textures, and the deep convolution layer gradually abstracts complex features including rock layer shape and crack direction. The improved activation function is improved on the basis of the traditional ReLU activation function, and the improved activation function expression is: ； In the above formula, It is a parameter between 0 and 1, and its value is adjusted according to the needs of geological feature identification; x It is the original feature value of each pixel in the feature map after the convolution operation. In the neural network, the role of the activation function is to perform nonlinear transformation on the feature map output by the convolution layer. The pooling layer uses maximum pooling or average pooling operations to perform dimensionality reduction processing on the feature map after convolution, thereby reducing the computational complexity, retaining the main geological feature information, and enhancing the robustness of the convolutional neural network to image scale changes; The fully connected layer is used to fully connect the pooled feature vector with the output of the convolutional neural network to map it to the target category corresponding to the geological feature; The output layer is used to output the final geological feature classification and recognition results, including category labels such as rock layers, cracks, and faults and their corresponding location information.

4. The method for three-dimensional visualization reconstruction of overlying rock mass structure of a deep coal mine working face according to claim 3 is characterized in that: The specific process of automatically extracting key geological features using the convolutional neural network in step S3 is as follows: Step S31, convolution operation; First, the parameters of the convolution kernel are initialized. The size, step length and number of the convolution kernel are determined according to the complexity of the geological features of the field exploration and the image resolution of the mining borehole imager. The original borehole image data converted into a numerical matrix is ​​input into the convolution kernel for convolution operation. Each pixel point of the image and the pixel value in its field are multiplied and summed with the corresponding weight of the convolution kernel to obtain the feature map after convolution. In the shallow convolution layer, basic features including the edge contour of the rock layer and the initial shape of the crack are extracted; the deep convolution layer extracts complex features including the layered structure of the rock layer, the extension direction and density of the crack. Step S32, applying a modified activation function to the convolution result; After the convolution operation in step S31, a modified activation function is applied to perform nonlinear transformation on the convolution result to enhance the learning and expression capabilities of the convolutional neural network; Step S33: pooling operation and full connection; After the nonlinear transformation, a pooling operation is performed. The maximum pooling or average pooling method is used to reduce the dimension of the convolution feature map according to a certain pooling window size and step size, thereby reducing the resolution of the feature map while retaining the main information of the geological features. The pooled feature vector is flattened and input into the fully connected layer, and the neurons of the fully connected layer are fully connected to each element of the flattened feature vector. Step S34, classify and output; According to the probability distribution of the output of the fully connected layer, the category with the largest probability is selected as the final geological feature classification result, and the classification result and the location information parameters of the corresponding geological features are output.

5. The method for three-dimensional visualization reconstruction of overlying rock mass structure of a deep coal mine working face according to claim 1, characterized in that: In step S4, the improved Kriging spatial interpolation method is used to build a three-dimensional geological model based on three-dimensional visualization software. The process is as follows: The traditional Kriging spatial interpolation method is used to calculate the weight coefficient The solution equations are transformed into nonlinear functions, and the weight coefficients are obtained by solving the nonlinear function : ； In the above formula, An unknown point in space and The variance function value between ; For sampling point and The variance function value between i =1,2,3,...,n, j =1,2,3,...,n, where n is the number of known sample points that can affect the sampling point; is the Lagrange multiplier; The calculation formula of the variance function value is: ； In the above formula, is the variation function value; h is the spatial distance; and The location and location The variable value at ; E is the symbol for variance calculation; The weight coefficient obtained Through equivalent infinitesimal transformation and Taylor formula expansion, we can get a high-order nonlinear function: ； The weight coefficient is obtained by differentiating the high-order nonlinear function The extreme value point, based on which the optimized weight coefficient is obtained , the optimized weight coefficient Substitute it into the basic equations of Kriging spatial interpolation method and deduce the weight coefficient The optimal value Z(x) of the unknown point is: ； The improved Kriging spatial interpolation method is used to perform linear unbiased optimal estimation of unknown points in the monitoring area through known sample points. Then, the discrete drilling data are interpolated layer by layer according to the divided strata, and the top and bottom surfaces of the strata are generated by three-dimensional visualization software. The surfaces are further combined into a body to construct a nonlinear three-dimensional geological model of the overlying rock mass structure of the deep coal mine working face.

6. The method for three-dimensional visualization reconstruction of overlying rock mass structure on a deep coal mine working face according to claim 1, characterized in that: In step S5, three-dimensional visualization of the overlying rock mass structure and risk areas of the deep coal mine working face is achieved through multimodal data fusion, and the process is as follows: Step S51, data preprocessing; The convolutional neural network is used to extract geological feature information including rock layers, fractures and faults from the exploration borehole images, and record their parameters including position, size and shape. At the same time, the introduced gas concentration and stress distribution data are normalized and converted into a numerical range suitable for the exploration borehole image data. Step S52: parameter quantization; The number of cracks per unit length or unit area in the original borehole image data is counted and recorded as For gas concentration, select the maximum gas concentration value in the monitoring area As a quantitative parameter; And determine the maximum stress value of the stress concentration point based on the stress distribution data And the area ratio of stress concentration area ; Step S53: constructing a comprehensive risk assessment algorithm model; The expression of the comprehensive risk assessment algorithm model is: ； In the above formula, represents the comprehensive risk value, , , They are the weight coefficients of the number of fractures, the maximum gas concentration and the maximum stress, which are determined according to the actual geological conditions and mining experience; the adjustment coefficient K=0.5; Considering the impact of the area ratio of stress concentration areas on risk, the comprehensive risk assessment algorithm model is modified to obtain the comprehensive risk value : ； In the above formula, k It is an adjustment coefficient used to regulate the influence of the proportion of stress concentration area on risk.

7. The method for three-dimensional visualization reconstruction of overlying rock mass structure on a deep coal mine working face according to claim 6, characterized in that: The overall risk value ≥0.8, the monitored area is a high-risk area; Comprehensive risk value 0.5 ≤ <0.8, the monitored area is a medium-risk area; the comprehensive risk value <0.5, the monitored area is a low-risk area.

Images