Advanced blasting pressure relief method based on rock burst prediction

Abstract

The invention discloses an advanced blasting pressure relief method based on rock burst prediction, and the method achieves the precise recognition and efficient intervention of rock burst risks through the combination of a micro-seismic monitoring technology, energy thermodynamic diagram modeling and an intelligent algorithm. According to the method, micro-seismic data is deeply analyzed, an abnormal activity area is identified, an energy thermodynamic diagram is constructed, a stress concentration situation is visually displayed, and risk level prediction is carried out in combination with historical data, so that early warning accuracy and response timeliness are improved, an optimal blast hole position in a high-risk area is automatically identified by adopting a density-based clustering algorithm, and the early warning efficiency is improved. Classification and directional blasting pressure relief are implemented according to risk levels, local stress is effectively released, the occurrence probability of rock burst is reduced, the overall process is highly automatic, monitoring, prediction, decision making and intervention are integrated, and the method is suitable for various complex coal mine geological environments.

Description

Technical Field

[0001] The present invention relates to the technical field of rock burst prevention and control, and in particular to an advance blasting pressure relief method based on rock burst prediction. Background Art

[0002] Rock burst is a common dynamic hazard in coal mining, primarily occurring in deep, highly stressed mining areas. Sudden rock instability causes the instantaneous release of massive amounts of energy, resulting in a violent outburst of coal and rock, accompanied by intense vibrations and posing a serious threat to miners and equipment. Its mechanism is complex, influenced by multiple factors, including geological structure, stress distribution, and mining methods. Preventing rock burst requires a comprehensive approach, including monitoring and early warning, regional governance, and on-site control.

[0003] However, during coal mining, conventional monitoring methods for rock burst mostly rely on a single data source and lack effective multi-dimensional data fusion methods, resulting in low positioning accuracy in microseismic anomaly areas. At the same time, there is a lack of in-depth analysis and refined processing of microseismic data, making it difficult to accurately identify potential risk areas for rock burst, limiting its application in high-precision early warning. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for pre-blasting pressure relief based on rock burst prediction to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for pre-blasting pressure relief based on rock burst prediction, comprising the following steps: S1. Microseismic detection nodes are deployed on the four structural surfaces of the coal mine roadway: the roof, left side, right side, and floor. This creates a closed distribution around the roadway cross section and extends forward in the direction of roadway excavation. These nodes are used to collect raw microseismic data, optimize the data, and extract short-time energy, spectral centroid, peak amplitude, signal kurtosis, and duration to construct a multidimensional feature vector. S2. Based on the constructed feature vector, an energy threshold judgment model is constructed to identify microseismic anomaly areas, abnormal nodes are calibrated by comparing with the set threshold, and the suspected earthquake source position is calculated based on the energy weighting of the abnormal node coordinates; S3. Based on the suspected earthquake source location and its energy value, a Gaussian diffusion function is used to perform energy simulation in three-dimensional space, and an energy distribution heat map is constructed to represent the local stress concentration area; S4. Use a spatiotemporal graph neural network risk prediction model to predict the probability of microseismic risk in the future, and dynamically construct warning levels based on the predicted probability. The warning levels are divided into three levels: low risk, medium risk, and high risk. S5. When the warning level is higher than low risk, the DBSCAN clustering algorithm is used to automatically locate the main pressure relief hole according to the stress concentration area in the thermal map. The blasting hole group is arranged with the main pressure relief hole as the center, combined with the coal mine roadway direction and stress direction. The blasting hole group is arranged on both sides of the main pressure relief hole in a fan-shaped array on both sides of the main pressure relief hole. S6. Initially detonate the blast hole cluster, leaving an interval of 30 to 60 minutes to partially fracture the coal structure. Then, place the water jet device into the main hole and perform a 10-20 minute hydraulic shock with a water jet pressure of 20 to 30 MPa. After the hydraulic shock is completed, detonate the blast hole cluster a second time, 10 to 15 minutes later, to expand the fracture zone. S7. Collect microseismic data in the area after blasting again and re-evaluate the local stress-energy state. If the energy drops below the set baseline within 24 hours, the intervention is considered successful. If the intervention is considered unsuccessful, further supplementary blasting or adjustment of the hole layout is performed.

[0006] Preferably, in step S1, the formula for extracting short-time energy is: in, is the original signal, is the sampling point index in the sliding window, For the signal at time energy density; The algorithm for extracting the spectrum centroid is: in, is the frequency variable, For signal The Fourier transform of For the signal at frequency The power spectral density at .

[0007] Preferably, in step S1, when performing data optimization processing on the original microseismic data, the data is subjected to wavelet transform denoising processing, and bandpass filtering is used to filter out The following low frequency noise and Above high frequency interference.

[0008] Preferably, in step S2, the energy threshold is the historical average energy value plus twice the standard deviation, and the suspected earthquake source position is obtained by calculating a weighted centroid positioning algorithm for multiple energy anomaly nodes.

[0009] Preferably, in step S3, the energy distribution thermodynamic map is generated by the following Gaussian diffusion function: ,in, is the source energy, is the coordinate of the earthquake source location, Indicates the target point arrive The square of the distance to the location of the earthquake source, is the diffusion radius.

[0010] Preferably, in step S4, the method for constructing the spatiotemporal graph neural network risk prediction model is: ,in, is the current graph structure, is the weight matrix corresponding to the graph structure, For historical time series characteristics, is the weight of the time series module, is the Sigmoid function, It is the model bias term, which is used to adjust the baseline offset of the prediction results.

[0011] Preferably, the current graph structure The construction method includes: taking multiple microseismic detection nodes as nodes of the graph, and based on the spatial distance between nodes Build adjacency relationships to meet At the node and Establish connections between them; Based on the characteristic sequence of microseismic events in the historical time window 、 , calculate the correlation weight The weighted adjacency matrix of the graph , and each node feature matrix Composition graph structure , the current graph structure The corresponding weight matrix It is a trainable parameter matrix for node feature transformation, acting on the node feature matrix .

[0012] Preferably, in step S5, the DBSCAN clustering algorithm is: ,in, is the heat map point set, is the neighborhood radius, The minimum number of points.

[0013] Preferably, in step S6, the diameter of the blasting hole group arranged on both sides of the main pressure relief hole is 90 mm, and the angle between the blasting hole group and the main pressure relief hole is ≥45°.

[0014] Preferably, in step S6, the diameter of the main pressure relief hole is 150 mm, the depth is 10-14 m, and the distance between each hole is 3-5 m.

[0015] By integrating microseismic data identification, energy thermogram modeling, risk prediction, and directional blasting intervention, this method accurately locates areas of microseismic anomalies. It also visually reflects stress concentration patterns through energy distribution maps and predicts risk levels at corresponding stress concentration locations, improving the accuracy and timeliness of early warnings. Furthermore, based on the DBSCAN algorithm, this method automatically locates blasthole locations and implements graded blasting interventions to effectively control rock burst risks. The overall process is highly automated and suitable for early warning and control of rock bursts in complex coal mine environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of the process of the advance blasting pressure relief method based on rock burst prediction according to an embodiment of the present invention; Figure 2 This is a thermal map showing a local stress concentration area in the advance blasting pressure relief method based on rock burst prediction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0018] See also Figure 1 The embodiment of the present invention provides a method for pre-blasting pressure relief based on rock burst prediction, comprising the following steps: S1. Microseismic detection nodes are arranged on the four structural surfaces of the coal mine roadway roof, left side, right side and bottom plate to form a closed distribution around the roadway cross section. The nodes are extended in front of the roadway excavation direction so that the spacing between nodes in each direction is 5m (specifically, the vertical spacing between the roof and bottom plate nodes is 5m, the horizontal spacing between the left and right side nodes is 5m, and the longitudinal spacing along the roadway direction is 5m). This forms a closed monitoring network. Each node is equipped with a wide-band acceleration sensor (frequency response range 0.1Hz-2kHz) to collect raw microseismic data. The collected raw microseismic data is optimized and processed, and the short-time energy, spectrum centroid, peak amplitude, signal kurtosis and duration are extracted to construct a multi-dimensional feature vector.

[0019] S11. In step S1, the formula for extracting short-time energy is: (Formula 1) Where, is the original signal, is the sampling point index in the sliding window, For the signal at time The energy density of the spectrum is obtained by using the sliding window method (window length 50ms, step length 25ms) and calculating the energy of each window according to the formula for extracting short-time energy. The algorithm for extracting the spectrum centroid is: (Formula 2) Where, is the frequency variable, For signal The Fourier transform of For the signal at frequency The power spectral density at .

[0020] S12: In step S1, when optimizing the original microseismic data, the data is subjected to wavelet transform denoising and bandpass filtering. The following low frequency noise and Above high frequency interference.

[0021] Furthermore, the multidimensional feature vector includes but is not limited to characteristic indicators such as short-time energy, spectrum centroid, peak amplitude, signal kurtosis, and duration. By splicing the features in multiple time windows, a multidimensional feature vector of fixed length is formed.

[0022] Specifically, in step S12, the original microseismic data is denoised by wavelet transform (using the db4 wavelet basis), and then a 10 Hz-1 kHz bandpass filter is used to eliminate low-frequency mechanical vibration and high-frequency electromagnetic interference.

[0023] S2. Based on the constructed eigenvector, an energy threshold judgment model is constructed to identify microseismic abnormal areas. Abnormal nodes are calibrated by comparing with the set threshold, and the suspected earthquake source location is calculated based on the energy weighting of the abnormal node coordinates.

[0024] S21, in step S2, the energy threshold is the historical average energy value plus twice the standard deviation, and the suspected earthquake source position is obtained by calculating the weighted centroid positioning algorithm for multiple energy anomaly nodes.

[0025] Among them, when calculating the source position through the weighted centroid positioning algorithm, the energy values ​​of abnormal nodes are logarithmically weighted to reduce the weight deviation of high energy points.

[0026] S3. Based on the suspected earthquake source location and its energy value, a Gaussian diffusion function is used to perform energy simulation in three-dimensional space, and an energy distribution heat map is constructed to represent the local stress concentration area.

[0027] S31. In step S3, the energy distribution heat map is generated by the following Gaussian diffusion function: (Formula 3) Where, is the source energy, is the coordinate of the earthquake source location, Indicates the target point arrive The square of the distance to the location of the earthquake source, is the diffusion radius.

[0028] Reference Figure 2 , select the diffusion radius Gaussian diffusion function, generating a three-dimensional energy heat map, by adjusting to , which can expand the display range of stress concentration areas and is suitable for large-scale stress monitoring scenarios.

[0029] S4. Use the spatiotemporal graph neural network risk prediction model to predict the probability of microseismic risk in the future, and dynamically construct warning levels based on the predicted probability. The warning levels are divided into three levels: low risk, medium risk, and high risk.

[0030] In step S41, the spatiotemporal graph neural network includes a temporal convolution layer, a graph convolution layer, and a fully connected layer. It uses a sliding window mechanism to input the graph sequence, uses a temporal convolution layer (TCN, convolution kernel size 5) to extract temporal features, and a graph convolution layer (GCN, where the adjacency matrix is ​​constructed based on the spatial distance of nodes) to extract spatial correlation. Finally, the features are fused through the fully connected layer to output the microseismic risk probability. .

[0031] In this embodiment, low risk: Only record data without triggering intervention, medium risk: Start local monitoring and trigger the blasting pressure relief process within the predetermined time. High risk: To immediately trigger the blasting pressure relief process.

[0032] S42. In step S4, the method for constructing a spatiotemporal graph neural network risk prediction model is: (Formula 4) Where, is the current graph structure, is the weight matrix corresponding to the graph structure, For historical time series characteristics, is the weight of the time series module, is the Sigmoid function, It is the model bias term, which is used to adjust the baseline offset of the prediction results.

[0033] In this embodiment, multiple microseismic detection nodes arranged in the coal mine tunnel are used as graph nodes of the graph neural network. The spatial position of each node can be obtained through the actual layout coordinates. According to the spatial distance between each node, , set the distance threshold ,like , then at node With node An undirected edge is established between them to reflect their spatial proximity relationship and obtain the initial adjacency matrix .

[0034] Furthermore, in order to improve the modeling ability of the graph structure for the correlation of monitoring data, the microseismic feature sequence of each node is extracted within the set time window. , including but not limited to microseismic amplitude, energy, frequency and other indicators, and using the Pearson correlation coefficient to calculate the feature similarity between each node pair, and weight the edges in the graph based on the similarity results to form a weighted adjacency matrix. The specific Pearson correlation coefficient formula is: (Formula 5) Where, Representation node In the historical time window The microseismic characteristic sequence within the adjacency matrix obtained by the above method and the corresponding node feature matrix Together they form a graph structure , for subsequent spatiotemporal graph neural network model to perform risk prediction modeling, the weight matrix corresponding to the graph structure It is a learnable parameter matrix in the graph neural network, used to adjust the node feature matrix Perform linear transformation and combine it with the normalized adjacency matrix , the graph structure features are extracted through graph convolution operation. The specific graph convolution formula is: (Formula 6) Where, is the activation function, The output graph structure is encoded for use by the subsequent time series modeling module. This method can effectively utilize the spatial structure information between microseismic monitoring points and improve the expressive ability of the risk prediction model.

[0035] S5. When the warning level is higher than low risk, the DBSCAN clustering algorithm is used to automatically locate the main pressure relief hole according to the stress concentration area in the thermal map. The blasting hole group is arranged with the main pressure relief hole as the center, combined with the direction of the coal mine tunnel and the stress direction. The blasting hole group is arranged on both sides of the main pressure relief hole and is arranged in a fan shape on both sides of the main pressure relief hole.

[0036] S51. In step S5, the DBSCAN clustering algorithm is: (Formula 7) Where, is the heat map point set, is the neighborhood radius, The minimum number of points.

[0037] S6. Detonate the blasting hole group for the first time, and wait 30 to 60 minutes after detonation to partially rupture the coal structure. Then place the water jet device into the main hole and perform water pressure shock for 10 to 20 minutes. The water jet pressure is 20 to 30 MPa. After the water pressure shock is completed, detonate the blasting hole group for the second time after an interval of 10 to 15 minutes to expand the fracture zone.

[0038] S61. In step S6, the diameter of the blasting hole group arranged on both sides of the main pressure relief hole is 90 mm, and the angle between the blasting hole and the main pressure relief hole is ≥45°.

[0039] S62. In step S6, the diameter of the main pressure relief hole is 150 mm, the depth is 10-14 m, and the distance between each hole is 3-5 m.

[0040] S7. Collect microseismic data in the area after blasting again and re-evaluate the local stress-energy state. If the energy drops below the set baseline within 24 hours, the intervention is considered successful. If the intervention is considered unsuccessful, further supplementary blasting or adjustment of the hole layout is performed.

[0041] In another embodiment, under hard rock geological conditions, the K-Means clustering algorithm is used instead of the DBSCAN clustering algorithm, and the number of clusters is set to , automatically locate the three main pressure relief holes in the stress concentration area. When arranging the blasting hole group, the fan-shaped angle on both sides of the main pressure relief holes is adjusted to 60°, and the hole spacing is increased to 2m.

[0042] In summary: By integrating microseismic data identification, energy thermodynamic map modeling, risk prediction and directional blasting intervention, accurate positioning of microseismic abnormal areas can be achieved, and the stress concentration situation can be intuitively reflected through the energy distribution map. The risk level can be predicted at the corresponding stress concentration location, thereby improving the accuracy and timeliness of early warning.

[0043] In addition, the DBSCAN algorithm is used to automatically locate blasting holes and implement graded blasting intervention to effectively control rock burst risks. The overall process has a high degree of automation and is suitable for early warning and control of rock burst in complex coal mine environments.

[0044] Parts not described in the present invention are the same as those in the prior art or can be implemented using the prior art. Although the embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for pre-blasting pressure relief based on rock burst prediction, characterized in that: The following steps are involved: S1. Microseismic detection nodes are deployed on the roof, left side, right side, and floor of the coal mine roadway, and are extended in the direction of tunneling. They are used to collect raw microseismic data, optimize the collected raw microseismic data, and extract the short-term energy, spectrum centroid, peak amplitude, signal kurtosis, and duration to construct a multidimensional feature vector. S2. Based on the constructed feature vector, an energy threshold judgment model is constructed to identify microseismic anomaly areas, abnormal nodes are calibrated by comparing with the set threshold, and the suspected earthquake source position is calculated based on the energy weighting of the abnormal node coordinates; S3. Based on the suspected earthquake source location and its energy value, a Gaussian diffusion function is used to perform energy simulation in three-dimensional space, and an energy distribution heat map is constructed to represent the local stress concentration area; S4. Use the spatiotemporal graph neural network risk prediction model to predict the probability of microseismic risk in the future and dynamically build an early warning level based on the predicted probability; S5. When the warning level shows medium or high risk, the DBSCAN clustering algorithm is used to automatically locate the main pressure relief hole according to the stress concentration area in the thermal map. The blasting hole group is arranged with the main pressure relief hole as the center, combined with the coal mine roadway direction and stress direction. The blasting hole group is arranged on both sides of the main pressure relief hole in a fan-shaped array on both sides of the main pressure relief hole. S6. Set up two blasting hole groups. After the first blast, perform water pressure shock on the holes. After the completion, perform a second blast to expand the fracture zone. S7. Collect microseismic data in the area after blasting again and re-evaluate the local stress-energy state. If the energy drops below the set baseline within 24 hours, the intervention is considered successful. If the intervention is considered unsuccessful, further supplementary blasting or adjustment of the hole layout is performed.

2. The method according to claim 1, wherein: In step S1, the formula for extracting short-time energy is: in, is the original signal, is the sampling point index in the sliding window, For the signal at time energy density; The algorithm for extracting the spectrum centroid is: in, is the frequency variable, For signal The Fourier transform of For the signal at frequency The power spectral density at .

3. The method according to claim 1, wherein: In step S1, when the original microseismic data is optimized, the data is subjected to wavelet transform denoising and bandpass filtering. The following low frequency noise and Above high frequency interference.

4. The method according to claim 1, wherein: In step S2, the energy threshold is the historical average energy value plus twice the standard deviation, and the suspected earthquake source position is obtained by calculating the weighted centroid positioning algorithm for multiple energy anomaly nodes.

5. The method according to claim 4, characterized in that: In step S3, the energy distribution heat map is generated by the following Gaussian diffusion function: ,in, is the source energy, is the coordinate of the earthquake source location, Indicates the target point arrive The square of the distance to the location of the earthquake source, is the diffusion radius.

6. The method according to claim 1, wherein: In step S4, the method for constructing the spatiotemporal graph neural network risk prediction model is: ,in, is the current graph structure, is the weight matrix corresponding to the graph structure, For historical time series characteristics, is the weight of the time series module, is the Sigmoid function.

7. The method according to claim 6, characterized in that: The current graph structure The construction method includes: taking multiple microseismic detection nodes as nodes of the graph, and based on the spatial distance between nodes Build adjacency relationships to meet At the node and Establishing connections between Based on the characteristic sequence of microseismic events in the historical time window 、 , calculate the correlation weight The weighted adjacency matrix of the graph , and each node feature matrix Composition graph structure , the current graph structure The corresponding weight matrix It is a trainable parameter matrix for node feature transformation, acting on the node feature matrix .

8. The method according to claim 7, wherein: In step S5, the DBSCAN clustering algorithm is: ,in, is the heat map point set, is the neighborhood radius, The minimum number of points.

9. The method according to claim 1, wherein: In step S6, the diameter of the blasting hole group arranged on both sides of the main pressure relief hole is 90 mm, and the angle between the blasting hole group and the main pressure relief hole is ≥45°.

10. The method for pre-blasting pressure relief based on rock burst prediction according to claim 1, characterized in that: In step S6, the diameter of the main pressure relief hole is 150 mm, the depth is 10-14 m, and the distance between each hole is 3-5 m.

Images