A DBSCAN-based method for identifying the boundaries of transient electromagnetic anomalies in mines.

Abstract

This invention belongs to the field of geophysical exploration technology and discloses a method for identifying the boundary of transient electromagnetic anomalies in mines based on DBSCAN clustering: Step 1, establish a three-dimensional geophysical model; Step 2, calculate apparent resistivity data using the small loop transient electromagnetic method; Step 3, draw a k-distance map and obtain the optimal neighborhood radius based on the k-distance map; Step 4, perform DBSCAN clustering on the apparent resistivity data to obtain the clustering results; Step 5, replace the apparent resistivity values ​​within each cluster with the cluster center values ​​to obtain the replaced apparent resistivity data; Step 6, image the replaced apparent resistivity data and identify anomalies; Step 7, compare the boundary of the anomaly with the boundary of the anomaly in the model. If the difference is too large, adjust the optimal neighborhood radius and return to Step 4; otherwise, use the current anomaly as the final identified anomaly. This invention can effectively identify the boundary of irregular anomalies in non-uniform and complex strata and can adapt to the real-time identification needs of underground anomalies.

Description

Technical Field

[0001] This invention belongs to the field of geophysical exploration technology, specifically involving a method for identifying the boundary of transient electromagnetic anomalies in mines based on DBSCAN clustering. This method is particularly suitable for the accurate identification of the boundaries of hidden geological anomalies such as coal mine goaf, faults, and collapse columns, and can also be extended to other geophysical imaging fields. Background Technology

[0002] Coal resources occupy an important position in my country's energy structure. During coal mining, hydrogeological hazards (such as goaf water accumulation, fault water conduction, and collapse column development) seriously threaten production safety. Accurate identification of the spatial distribution and boundary characteristics of hidden anomalies is crucial to guide disaster prevention and engineering decisions. The transient electromagnetic method in mines has good detection sensitivity for low-resistivity thin layers and water-rich bodies, making it an important means of detecting mine water hazards. However, it suffers from fuzzy boundary identification of anomalies under complex geological conditions. Traditional forward and inverse methods are constrained by stratigraphic continuity, and the gradual changes in apparent resistivity data lead to unclear anomaly boundary characterization, especially for irregular boundaries in non-homogeneous strata. Existing technologies, derivative-based boundary identification methods are insufficiently adaptable to complex anomaly shapes, and the application of machine learning methods in geophysical data interpretation still needs optimization. Therefore, a technical solution that can effectively improve the accuracy of anomaly boundary identification is urgently needed.

[0003] Clustering analysis, as a data-driven unsupervised learning method, is mainly used for pattern recognition, feature classification, and anomaly segmentation in geophysical data processing. However, there is currently no systematic method for applying the DBSCAN clustering algorithm to the boundary identification of transient electromagnetic anomalies in mines. Transient electromagnetic data is affected by diffusion effects, exhibiting gradual boundary characteristics and strong noise interference. Traditional clustering methods (such as K-means) require a pre-defined number of clusters and cannot handle irregular anomalies; while existing DBSCAN applications (such as slope monitoring) rely on iterative optimization, resulting in low computational efficiency and unsuitability for the real-time identification of underground anomalies. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying the boundary of transient electromagnetic anomalies in mines based on DBSCAN clustering, so as to solve the problems that existing clustering methods cannot handle irregular anomalies and that existing DBSCAN application technology is not suitable for the real-time identification of underground anomalies.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for identifying the boundaries of transient electromagnetic anomalies in mines based on DBSCAN clustering includes the following steps:

[0007] Step 1: Use 3D modeling software to create a 3D geophysical model of the mining area to be identified.

[0008] Step 2: Perform forward modeling on the three-dimensional geophysical model obtained in Step 1 using the small loop transient electromagnetic method to obtain apparent resistivity data.

[0009] Step 3: Use the apparent resistivity data obtained in Step 2 to plot a k-distance map, and obtain the optimal neighborhood radius based on the k-distance map.

[0010] k-distance refers to the maximum distance from a data point to its k nearest neighbors. As the distance increases, the k-distance map usually shows an increasing trend until the distance to a certain point suddenly increases significantly, forming a clear inflection point. This inflection point is the optimal neighborhood radius, which can help distinguish core points from noise points.

[0011] Step 4: Set the minimum number of points; perform DBSCAN clustering on the apparent resistivity data based on the optimal neighborhood radius and the minimum number of points to obtain the clustering results, including several clusters and noise points, and delete the noise points.

[0012] Specifically, in DBSCAN clustering, the apparent resistivity data obtained in step 2... any point in the middle The neighborhood radius is defined as follows:

[0013]

[0014] In the formula:

[0015] — Apparent resistivity data x neighborhood radius All points within;

[0016] , —Any two distinct points in the apparent resistivity data;

[0017] -point With point Euclidean distance;

[0018] —Optimal neighborhood radius;

[0019] Step 5: Replace the apparent resistivity value in each cluster with the apparent resistivity value of the cluster center point, while keeping the coordinates unchanged, to obtain the replaced apparent resistivity data.

[0020] Step 6: Image the replaced apparent resistivity data and identify the boundary of the anomalous body from the imaging results to obtain the anomalous body;

[0021] Step 7: Compare the extracted anomaly boundary with the anomaly boundary in the 3D geophysical model established in Step 1. If the difference is too large, adjust the current optimal neighborhood radius (preferably ±0.5) and return to Step 4 for iterative iteration; otherwise, take the clustering result of Step 4 in the current iteration as the optimal clustering result. The anomaly obtained in Step 6 is taken as the final identified anomaly.

[0022] Furthermore, in step 1, the 3D modeling software used is Blender.

[0023] Furthermore, in step 4, the minimum number of points is equal to the array dimension of the apparent resistivity data plus 1.

[0024] Furthermore, in step 7, the "excessive difference" specifically refers to the area within the boundary of the extracted anomaly being two times or more different from the area within the boundary of the anomaly in the three-dimensional geophysical model obtained in step 1.

[0025] Compared with the prior art, the present invention has the following technical effects:

[0026] (1) The method of the present invention effectively distinguishes the background field from the abnormal field by aggregating the apparent resistivity features of the multidimensional feature space, enabling DBSCAN clustering to directly capture the geometric distribution of the abnormal body and improve the ability to identify low signal-to-noise ratio data. Therefore, there is no need to preset the shape and number of the abnormal body, which is suitable for discovering any irregularly shaped abnormal body and reducing manual intervention.

[0027] (2) It is feasible in engineering, and the post-processing is based on the apparent resistivity results. The amount of calculation is moderate and it supports automated deployment.

[0028] (3) This invention utilizes the inflection point of the k-distance curve to quickly determine the optimal neighborhood radius, eliminating the need for iterative optimization and adapting to the real-time processing needs of mines.

[0029] (4) By replacing the intra-cluster resistivity with the central value and filtering noise, the limitations of traditional imaging constrained by the continuity of strata are overcome, the boundaries are significantly sharpened, the problem of boundary blurring under complex geological conditions is solved, and the sharpening and identification of anomalous boundaries is achieved. It can be applied to multiple scenarios such as low resistivity, high resistivity and complex anomalous bodies, and has promotional value. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the goaf model in Example 1;

[0031] Figure 2 This is a cross-sectional view of the apparent resistivity of the goaf in Example 1;

[0032] Figure 3 This is a k-distance map of the goaf data from Example 1;

[0033] Figure 4 This is the apparent resistivity contour map after DBSCAN clustering replacement in Example 1;

[0034] Figure 5 This is a graph of the K-means elbow method from Example 1;

[0035] Figure 6 This is the apparent resistivity contour map after K-means clustering replacement in Example 1;

[0036] Figure 7 This is a schematic diagram of the fault anomaly model in Example 2;

[0037] Figure 8 This is an apparent resistivity profile of the fault anomaly model in Example 2;

[0038] Figure 9 This is a k-distance map of fault data from Example 2;

[0039] Figure 10 This is the tomographic resistivity contour map after DBSCAN clustering replacement in Example 2;

[0040] Figure 11 This is a schematic diagram of the composite model of coal seam collapse column and water-filled roadway in Example 3;

[0041] Figure 12 This is a cross-sectional view of the apparent resistivity of the composite model in Example 3;

[0042] Figure 13 This is the k-distance map of the composite model data in Example 3;

[0043] Figure 14 The images show the apparent resistivity contour maps after clustering replacement in Example 3 using the composite model. In (a), the optimal neighborhood radius is 9.6, and in (b), the optimal neighborhood radius is 13. Detailed Implementation

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments, but the implementation of the present invention is not limited thereto. Specific implementation steps are as follows:

[0045] Example 1: Identification of Abnormal Boundaries in Goaf Areas

[0046] Establish as Figure 1 The cubic low-resistivity water-filled goaf model shown has a low-resistivity anomaly center buried at a depth of 40m, dimensions of 20m × 20m × 20m, and a resistivity of 10. The resistivity of the overlying and underlying strata is 500. The resistivity of the space where the tunnel is located is 1.0 × 10⁻⁶. 8 Forward modeling was performed using the small loop transient electromagnetic method. The loop was placed in the tunnel space along the x-direction, with a loop size of 3m × 3m, a measurement point interval of 3m, a transmission current of 10A, an acquisition time of 1.0 × 10⁻⁷ to 1.0 × 10⁻³ s, a measurement range of -88.5 to 88.5m, and 100 time channels.

[0047] Forward modeling was performed on the model. For numerical simulation, the finite element method (FEM) was used, and the mesh was generated using the Tetgen open-source software. During mesh generation, appropriate mesh sizes and radius-to-edge ratios were selected based on the complexity of different models and computational requirements, balancing computational accuracy with efficiency. Forward calculations were performed on the range of -88.5 to 88.5 m, and the apparent resistivity profile (at y=0) was obtained, as shown below. Figure 2 As shown.

[0048] Figure 2 In the figure, there is a significant low-resistivity anomaly response at a depth of 40m underground. The blue and green areas in the figure, and the area highlighted in red, represent the actual area of ​​the anomaly. If the blue area is interpreted as the anomaly range, the ratio of the area of ​​the blue area to the area of ​​the actual anomaly is 2.82 / 2.31 = 1.22. If the blue and green areas are interpreted as the anomaly range, the ratio of its area to the actual anomaly area is 7.6 / 2.31 = 3.29. However, in the interpretation of geophysical anomalies, the anomaly range is often interpreted as the blue and green areas.

[0049] DBSCAN clustering is performed on the apparent resistivity data. For the apparent resistivity data of goaf anomalies, the minimum number of points for DBSCAN clustering is set to minPts=3 (generally the data dimension + 1). Based on the k-distance map of the goaf data, as shown... Figure 3 As shown in the figure, the inflection point value observed is around 8, so a DBSCAN clustering method with an optimal neighborhood radius of 8 is selected. The apparent resistivity of each cluster is then replaced with the apparent resistivity value of the cluster's center point, while maintaining the coordinates. The replaced apparent resistivity data is then imaged as follows: Figure 4 As shown. Figure 4 The black box indicates the anomaly range interpreted by DBSCAN clustering, while the red box indicates the actual anomaly range. The area ratio of the two is 2.37 / 2.31 = 1.02. The DBSCAN clustering algorithm demonstrates excellent recognition ability for apparent resistivity anomaly boundaries, with significant boundary sharpening effect, clearly outlining the geometric features of the cubic anomaly.

[0050] As a comparison, K-means clustering was performed on the apparent resistivity data. For the apparent resistivity data of anomalies in the goaf, the optimal number of clusters for the goaf data was determined using the elbow method. Figure 5As shown in the figure, the optimal number of clusters observed is 4. K-means clustering is performed, and the apparent resistivity image after clustering is as follows. Figure 6 As shown. Figure 6 The black box indicates the anomaly range explained by K-means clustering, while the red box indicates the actual anomaly range. The area ratio of the two is 3.37 / 2.31 = 1.46. This shows that K-means clustering also demonstrates excellent ability to identify the boundaries of resistivity anomalies, but it is slightly inferior to DBSCAN clustering.

[0051] Example 2: Identification of Fault Anomalies

[0052] Establish as Figure 7 The high-resistivity layered fault anomaly model shown simulates a displaced coal seam, with dimensions of 400m × 100m × 20m. The center of the hanging wall is 30m above the roadway surface, and the center of the footwall is 40m above the roadway surface. The fault thickness is 20m, and the coal seam resistivity is 500 Ω·cm. The resistivity of the overlying and underlying strata is 100. The resistivity of the space where the tunnel is located is 1.0 × 10⁻⁶. 8 Forward modeling was performed using the small loop transient electromagnetic method. The loop was placed in the tunnel space along the x-direction, with a loop size of 3m × 3m, a measurement point interval of 3m, a transmission current of 10A, an acquisition time of 1.0 × 10⁻⁷ to 1.0 × 10⁻³ s, a measurement range of -199.5 to 199.5m, and 100 time channels.

[0053] Forward modeling was performed on the model. For numerical simulation, the finite element method (FEM) was used, and the mesh was generated using the Tetgen open-source software. During mesh generation, appropriate mesh sizes and radius-to-edge ratios were selected based on the complexity of different models and computational requirements to balance computational accuracy with efficiency. Forward calculations were performed on the range of -199.5 to 199.5, and the apparent resistivity profile (at y=0) was obtained, as shown below. Figure 8 As shown in the figure. The area highlighted in black in the figure is the actual location of the fault.

[0054] Similar to Example 1, K-distance plots were plotted on the apparent resistivity data obtained from the forward modeling of the high blocking layer model, as shown below. Figure 9 As shown, the inflection point value observed from the k-distance plot is between 3 and 4. Therefore, the optimal neighborhood radius = 3.7 and the minimum number of points minPts = 3 are selected for DBSCAN clustering. The apparent resistivity of each cluster is then replaced with the apparent resistivity value of the cluster's center point, while maintaining the coordinates. The replaced apparent resistivity values ​​are then imaged as shown below. Figure 10 As shown, Figure 10The area marked by the black rectangle in the middle represents the actual anomaly zone of the fault. Analysis shows that the hierarchical clustering method has a significant ability to identify the boundaries of the hanging wall and footwall of the fault, and its anomaly boundaries are highly consistent with the resistivity gradient variation characteristics.

[0055] Example 3: Identification of composite anomalies of coal seam collapse columns and water-filled roadways

[0056] Establish as Figure 11 The model shown is a composite anomaly of a coal seam water-filled collapse column and a water-filled roadway. The coal seam dimensions are 400m × 100m × 30m, the center is 55m from the roadway surface, the fault thickness is 30m, and the coal seam resistivity is 500 Ω·cm. The resistivity of the overlying and underlying strata is 100. The resistivity of the space where the tunnel is located is 1.0 × 10⁻⁶. 8 In the middle of the coal seam, at a horizontal position x=-100, there is a low-resistivity spherical collapse column with a radius of 25m and a center position 55m above the roadway surface. Its resistivity is 10⁻⁶. There is a water-filled roadway with dimensions of 10m × 100m × 10m in the upper right part of the coal seam, with a horizontal position x=50. Its center is 40m above the roadway surface, and its resistivity is 5. .

[0057] Forward modeling was performed using the small loop transient electromagnetic method. The loop was placed in the tunnel space along the x-direction, with a loop size of 3m × 3m, a measurement point interval of 3m, a transmission current of 10A, a acquisition time of 1.0 × 10⁻⁷ to 1.0 × 10⁻³ s, a measurement range of -229.5m to 118.5m, and 100 time channels. Forward modeling was performed using the finite element method (FEM) for numerical simulation, with mesh generation using the Tetgen open-source software. During mesh generation, appropriate mesh sizes and radius-to-edge ratios were selected based on the complexity of different models and computational requirements to balance computational accuracy with efficiency. Forward calculations were performed on the range of -229.5m to 118.5m, and the apparent resistivity profile (at y=0) was obtained, as shown below. Figure 12 As shown in the figure. The area highlighted in black in the figure is the actual location of the fault.

[0058] Similarly, K-distance plots are drawn from the apparent resistivity data obtained by forward modeling of the composite model, as shown above. Figure 13 As shown, the inflection point values ​​observed from the k-distance plot are between 8 and 14. Therefore, DBSCAN clustering was performed with optimal neighborhood radii of 13 and 9.6, and a minimum number of points minPts=3. The apparent resistivity images after clustering are then displayed as shown in the figure. Figure 14 As shown, Figure 14The black-marked areas in the upper and lower sub-figures represent the actual anomaly regions of the composite model. When the optimal neighborhood radius is 13, the identification of collapse columns and water-filled roadways is obvious, but the coal seam is not identified. When the optimal neighborhood radius is 9.6, the collapse columns and water-filled roadways are accurately identified, and the coal seam boundary is also clearly delineated. It can be seen that the method of this invention can effectively identify the composite anomaly of coal seam collapse columns and water-filled roadways.

[0059] The above are preferred embodiments, but the present invention is not limited to the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution, and these simple modifications all fall within the scope defined by the claims of this application.

[0060] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable way without contradiction. In order to avoid unnecessary repetition, the various possible combinations will not be described separately here.

[0061] Furthermore, various implementation methods of this embodiment can be combined arbitrarily, as long as they do not violate the spirit of the technical solution of this invention, and should also be considered within the scope defined by the claims of this application.

Claims

1. A method for identifying the boundaries of transient electromagnetic anomalies in mines based on DBSCAN clustering, characterized in that, Specifically, the following steps are included: Step 1: Use 3D modeling software to create a 3D geophysical model of the mine area to be identified; Step 2: Perform forward modeling on the three-dimensional geophysical model obtained in Step 1 using the small loop transient electromagnetic method to obtain apparent resistivity data. Step 3: Use the apparent resistivity data obtained in Step 2 to plot a k-distance map, and obtain the optimal neighborhood radius based on the k-distance map; Step 4: Set the minimum number of points, and perform DBSCAN clustering on the apparent resistivity data based on the optimal neighborhood radius and the minimum number of points to obtain the clustering results, including several clusters and noise points, and delete the noise points; Step 5: Replace the apparent resistivity value in each cluster with the apparent resistivity value of the cluster center point, while keeping the coordinates unchanged, to obtain the replaced apparent resistivity data; Step 6: Image the replaced apparent resistivity data and identify the boundary of the anomalous body from the imaging results. Step 7: Compare the extracted anomaly boundary with the anomaly boundary in the three-dimensional geophysical model established in Step 1. If the difference between the two is too large, adjust the current optimal neighborhood radius and return to Step 4 for iterative iteration; otherwise, take the anomaly obtained in Step 6 in the current iteration as the final identified anomaly.

2. The method for identifying the boundary of transient electromagnetic anomalies in mines based on DBSCAN clustering as described in claim 1, characterized in that, In step 1, the 3D modeling software used is Blender.

3. The method for identifying the boundary of transient electromagnetic anomalies in mines based on DBSCAN clustering as described in claim 1, characterized in that, In step 4, the minimum number of points is equal to the array dimension of the apparent resistivity data plus 1.

4. The method for identifying the boundary of transient electromagnetic anomalies in mines based on DBSCAN clustering as described in claim 1, characterized in that, In step 7, the "excessive difference" specifically refers to the area within the boundary of the extracted anomaly being two times or more different from the area within the boundary of the anomaly in the three-dimensional geophysical model obtained in step 1.

Images