Mine earthquake directory construction method and system based on dual-drive classification inversion

Abstract

The invention provides a mine earthquake directory construction method and system based on dual-drive classification inversion, and belongs to the technical field of mine earthquake monitoring. Comprising the following steps: acquiring real-time data during mine vibration and mining activity; constructing a mine noise fingerprint database, and performing scene adaptive noise filtering on the obtained data; performing event type division on the de-noised data based on a dual-channel CNN architecture, and performing classification inversion according to an event type division result to realize focus parameter optimization; and fusing the event type division result and the seismic source parameters to generate a mine seismic directory. According to the invention, through dual-drive and classification inversion of physical data and deep learning data, seismic directory dynamic construction with high positioning precision can be realized in a complex mine noise environment.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of mine earthquake monitoring, and particularly relates to a mine earthquake catalog construction method and system based on double-drive classification inversion. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] Mine earthquake is one of the major hidden dangers of mine safety production, which is strong in suddenness under complex mining environment, easy to cause sudden rupture of rock mass, induce rock burst, lead to underground roadway collapse, and directly threaten the safety of workers' lives, and at the same time, damage mining equipment and interrupt production process. Moreover, due to the large noise interference in mine and the high monitoring difficulty, the problems of early warning lag and positioning error further aggravate the disaster risk. Therefore, it is of great value to construct a mine earthquake catalog.

[0004] However, the existing mine earthquake catalog construction method has some difficult-to-solve technical problems, for example: (1) In the existing technology, the classification of mine earthquake related events mainly relies on a single physical feature (such as P / S wave amplitude ratio) to divide event types, which is difficult to adapt to complex working conditions, and thus leads to many mine earthquakes in coal mines being often misjudged as blasting events; some existing machine learning models can only rely on waveform statistical features, which cannot deeply analyze the physical mechanism of the source.

[0005] (2) The noise environment of mine earthquake monitoring area is complex, and the existing technology lacks a dedicated noise feature system for mine scene, which makes it difficult to reasonably process each denoising stage, for example: there are residual strong interference signals, or the details of microseisms are lost due to excessive filtering. At the same time, the denoising model is trained based on general data, which is not adapted to the non-stationary characteristics of mine noise, and the denoising effect in complex areas is obviously attenuated.

[0006] (3) In the process of collecting related data, the fixed frequency response range of the seismic sensor in the existing technology cannot capture the extremely low frequency surface wave and high frequency microseismic signal; when using a unified sampling rate, the waveform of a large energy event is easily distorted due to clipping, and a small magnitude event is easily missed due to insufficient signal-to-noise ratio. SUMMARY

[0007] In order to overcome the above-mentioned deficiencies of the prior art, the present application provides a mine earthquake catalog construction method and system based on double-drive classification inversion, which can realize dynamic construction of earthquake catalog with high positioning accuracy in complex mine noise environment through double-drive and classification inversion of physical data and deep learning data.

[0008] To achieve the above object, one or more embodiments of the present application provide the following technical solutions: The first aspect of the present application provides a mine seismic catalog construction method based on double-drive classification inversion.

[0009] A mine seismic catalog construction method based on double-drive classification inversion comprises: Real-time data during mine vibration and mining activities are collected by a multi-source sensor network, and energy detection is performed; the multi-source sensor network automatically switches the collection mode according to the energy detection result; A mine noise fingerprint library is constructed, and the obtained real-time data is subjected to scene adaptive noise filtering; The data after noise reduction is subjected to event type division based on a double-channel CNN architecture; According to the event type division result, classification inversion is performed to realize optimization of the source parameters, specifically: when the event type is divided into a mine seismic event, moment tensor inversion is performed; when the event type is divided into blasting or noise, basic positioning inversion is performed; and a mine seismic catalog is generated by fusing the event type division result and the source parameters.

[0010] Further, the multi-source sensor network is a surface- underground three-dimensional monitoring network composed of surface wave sensors, wideband accelerometers and distributed optical fiber sensors.

[0011] Further, the multi-source sensor network automatically switches the collection mode according to the energy detection result, comprising: when a large energy event is detected, automatically switching to a down-sampling processing mode; and when a micro energy event is detected, automatically switching to an up-gain processing mode.

[0012] Further, the scene adaptive noise filtering comprises: strong filtering during the blasting period based on the mining plan, noise suppression of the data during the non-blasting period by using a transfer learning denoising model, and quality grading control of the filtered data according to a dynamic signal-to-noise ratio threshold.

[0013] Further, the double-channel CNN architecture comprises: a first channel, a second channel, a fusion layer, a full connection layer and a classifier; wherein the features of the first channel and the second channel are first spliced with physical features through the fusion layer, and then output the event type probability through the full connection layer and the classifier.

[0014] Further, the first channel uses 1D-CNN to process the original waveform segment of the collected data, and the second channel uses 2D-CNN to process the time-frequency graph generated by short-time Fourier transform.

[0015] Further, when the event type is divided into mine earthquake events, the moment tensor inversion is performed, including: determining the level of the mine earthquake event, when the belonging level is a significant mine earthquake, the moment tensor inversion is used for positioning; when the belonging level is a micro mine earthquake, the waveform cross-correlation and array beam forming technology are used for positioning; when the belonging level is a very micro mine earthquake, the detection rate is improved by optimizing the array configuration.

[0016] The second aspect of the present application provides a mine earthquake catalog construction system based on double-drive classification inversion.

[0017] A mine earthquake catalog construction system based on double-drive classification inversion comprises: A multi-source sensor network module is configured to collect real-time data during mine vibration and mining activities, and perform energy detection; at the same time, the collection mode is automatically switched according to the energy detection result; A scene adaptive noise filtering module is configured to construct a mine noise fingerprint library, and perform scene adaptive noise filtering on the obtained real-time data; An event type division module is configured to divide the event type of the de-noised data based on a double-channel CNN architecture; A classification inversion module is configured to perform classification inversion according to the event type division result, realize source parameter optimization, and specifically: when the event type is divided into mine earthquake events, the moment tensor inversion is performed; when the event type is divided into blasting or noise, the basic positioning inversion is performed; and the mine earthquake catalog is generated by fusing the event type division result and the source parameter. The third aspect of the present application provides a computer readable storage medium, which stores a program, and the program is executed by a processor to realize the steps of the mine earthquake catalog construction method based on double-drive classification inversion according to the first aspect of the present application.

[0018] The fourth aspect of the present application provides an electronic device, which comprises a memory, a processor and a program stored in the memory and executable on the processor, and the processor executes the program to realize the steps of the mine earthquake catalog construction method based on double-drive classification inversion according to the first aspect of the present application.

[0019] The above one or more technical solutions have the following beneficial effects: (1) The application realizes accurate division of event types by adopting an event classification scheme of "physical-data double driving", based on a double-channel CNN architecture (the first channel processes the original waveform segment, and the second channel processes the time-frequency graph), and splicing physical characteristics (such as P / S wave spectrum ratio, matrix tensor anisotropy index) in the fusion layer, effectively reducing the misjudgment rate of mine earthquake and blasting events; at the same time, by fusing physical characteristics and deep learning characteristics, the source physical mechanism can be deeply analyzed, and the limitation of relying on single feature or pure data driven model in the prior art is broken through.

[0020] (2) The application realizes reasonable processing of each denoising stage by constructing a mine noise fingerprint library, combining a scene adaptive noise filtering strategy (strong filtering during blasting period, denoising model of transfer learning during non-blasting period, dynamic signal-to-noise ratio threshold quality control), which not only avoids the residual of strong interference signals, but also prevents the loss of microseismic details caused by excessive filtering; at the same time, the denoising model of transfer learning can adapt to the non-stationary characteristics of mine noise, and significantly improves the denoising effect in complex areas.

[0021] (3) The application realizes comprehensive acquisition of very low frequency surface wave and high frequency microseismic signal by deploying a multi-source sensor network composed of surface wave sensors, wideband accelerometers and distributed optical fiber sensors; at the same time, the sensor network can automatically switch the acquisition mode (downsampling for high energy events, upgain for micro energy events) according to the energy detection result, effectively avoiding the waveform distortion caused by clipping of high energy events, improving the detection rate of small magnitude events, and reducing the missed detection range.

[0022] The advantages of the additional aspects of the application will be partially given in the following description, partially become obvious from the following description, or be understood by the practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0023] The drawings accompanying the specification of this application form a part thereof, serve to provide further understanding of the application, and together with the description of the exemplary embodiments of the application and explanations thereof serve to explain the application, and do not constitute improper limitations on the application.

[0024] Figure 1 A flow chart of a mine earthquake catalog construction method based on double driving classification inversion in the embodiment one of the application. DETAILED DESCRIPTION

[0025] It should be noted that the following detailed description is exemplary, and is intended to provide further explanation of the application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the application belongs.

[0026] It is to be noted that the terms used herein are merely for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application.

[0027] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0028] Embodiment one The embodiment discloses a mine earthquake catalog construction method based on double-drive classification inversion.

[0029] As Figure 1 shown, a mine earthquake catalog construction method based on double-drive classification inversion comprises the following steps: Step S1, real-time data of mine vibration and mining activities are collected by a multi-source sensor network, and energy detection is performed; the multi-source sensor network automatically switches the collection mode according to the energy detection result; Step S2, a mine noise fingerprint library is constructed, and the obtained real-time data is subjected to scene adaptive noise filtering; Step S3, the data after noise reduction is subjected to event type division based on a double-channel CNN architecture; Step S4, classification inversion is performed according to the event type division result, and source parameter optimization is realized, specifically: when the event type division is a mine earthquake event, moment tensor inversion is performed; when the event type division is blasting or noise, basic positioning inversion is performed; and a mine earthquake catalog is generated by fusing the event type division result and the source parameter.

[0030] Based on the above process, the present application can realize dynamic construction of an earthquake catalog with high positioning accuracy in a complex mine noise environment through double driving of physical data and deep learning data and classification inversion. In order to facilitate the understanding of the technical scheme of the present application, the specific implementation method in the technical scheme of the present application will be further explained and described below.

[0031] In step S1, real-time data of mine vibration and mining activities are collected by a multi-source sensor network, and energy detection is performed.

[0032] A multi-source sensor network for real-time data collection is established by deploying a multi-modal sensor array, and the multi-source sensor network is a surface-underground three-dimensional monitoring network composed of surface wave sensors, wideband accelerometers and distributed optical fiber sensors. Specifically: 1) the installation positions of the surface wave sensors, wideband accelerometers and distributed optical fiber sensors are as follows: The surface wave sensors are directly fixed to the shallow layer (floor or sidewall, depth < 3 m) of the roadway or the boundary of the goaf, and are tightly attached to the rock surface through adhesive or magnetic seat, so as to capture the surface wave signals of shallow crack propagation.

[0033] Broadband accelerometers are mainly deployed in deep boreholes (depth between 7-10 m, penetrating the loose circle of the roadway) to ensure close coupling with stable rock mass; at the same time, they are also laid in high stress areas (such as fault zones, coal pillars) and key equipment bases (coal mining machine bearing seat) to capture high-frequency signals of rock mass rupture and mechanical vibration anomalies.

[0034] Distributed acoustic sensors (DAS) are mainly laid along the sidewall of the roadway or cable trough (such as transportation roadway, return airway); or implanted in the ground, underground boreholes (vertical holes to the upper part of the unmined coal seam, near-horizontal holes in hard rock), to monitor large-scale strain field; they can also be wound around support structures (such as grouting steel tanks) to realize real-time sensing of support system strain migration.

[0035] 2) The actual role of surface wave sensors, broadband accelerometers and distributed acoustic sensors is as follows: surface wave sensors are mainly used to capture shallow rock mass surface waves to assist in analyzing rock mass rupture propagation; broadband accelerometers are mainly used to capture high-frequency vibration signals (P waves, S waves) generated by rock mass rupture to provide early warning of disasters such as rock burst; distributed acoustic sensors (DAS) realize long-distance, dense spatial sampling by sensing strain rate through optical fibers, suitable for large-scale microseismic wave field monitoring.

[0036] 3) The working process of surface wave sensors, broadband accelerometers and distributed acoustic sensors when collecting real-time data during mine vibration and mining activities, and when detecting energy, is as follows: Surface wave sensors work based on the principle of surface acoustic wave (SAW), using the piezoelectric effect to convert mechanical vibration into electrical signals. When mine vibration occurs, vibration waves will affect the propagation characteristics (such as frequency or phase) of surface acoustic waves on the piezoelectric substrate, and the system detects these changes to sense vibration energy.

[0037] Broadband accelerometers sense vibration acceleration through an internal mass. When mine vibration occurs, the mass will produce displacement, which is converted into an electrical signal through piezoelectric or capacitive effect, thereby detecting vibration energy. Its characteristic is that it can capture vibration signals in a wide frequency range (from several hertz to several kilohertz).

[0038] Distributed acoustic sensors (DAS) use the backscattering effect (such as Rayleigh scattering, Brillouin scattering) of light when it propagates in optical fibers. Mine vibration will cause strain or temperature changes in the optical fiber, changing the characteristics of scattered light, and the system can realize long-distance, continuous spatial vibration energy detection by analyzing these changes.

[0039] On this basis, the multi-source sensor network adopts a dynamic configuration strategy to automatically switch the collection mode according to the energy detection results. The dynamic configuration strategy runs on the edge computing node or the central server, and dynamically adjusts the working parameters and network topology of the sensor by real-time analysis of multi-source data (such as vibration signals, strain changes, and environmental noise). A single sensor only has basic data collection and local preprocessing capabilities (such as filtering and compression), while the decision logic of the dynamic strategy (such as task allocation and wake-up mechanism) is executed by the central system. According to the event priority (such as rock burst warning), the accelerometer in sleep mode is automatically awakened, and the sampling rate of DAS is increased. During the low-risk period, redundant sensors are turned off or power consumption is reduced (such as DAS entering low-power mode). The communication path is dynamically adjusted by a self-organizing algorithm. For example, when the DAS optical cable in a certain area is broken, the system automatically switches to the adjacent accelerometer as a relay node to maintain data transmission. Based on the real-time signal-to-noise ratio, the gain of the surface wave sensor or the frequency band range of the accelerometer is dynamically adjusted to suppress environmental interference. The specific implementation includes: 1) Scene-driven: If the DAS detects abnormal strain in the goaf (such as separation settlement), the system automatically schedules nearby accelerometers to focus on high-frequency vibration monitoring in that area, and activates surface wave sensors to analyze shallow fracture expansion.

[0040] 2) Data fusion feedback: The central system analyzes multi-source data (such as vibration waveforms and optical fiber strain mapping) through a deep learning model to real-time correct sensor deployment density or location (such as adding mobile sensor nodes to high-stress areas).

[0041] Therefore, based on the above design, the following operations can be performed: when a large energy event is detected, automatically switch to a down-sampling processing mode, such as automatically switching to a 2000Hz sampling rate and a 20dB gain; when a micro-energy event is detected, automatically switch to an up-gain processing mode, such as automatically switching to a 5000Hz sampling rate and a 60dB gain high-sensitivity mode. As an optional embodiment, an energy greater than 1000 joules can be considered as a large energy event, and vice versa.

[0042] In order to further guarantee the quality of the data collected by the multi-source sensor network, a sensor health degree evaluation model is constructed for evaluating the multi-source sensor network, that is: ; Among them, represents the comprehensive health degree (0-1, 1 represents complete health); 、 and represent weight coefficients, and can be adjusted according to the application scenario (such as mine earthquake monitoring focusing on dynamic response, then is higher). denotes the Bayesian-based node reliability, denotes the dynamic property health, denotes the data-driven health.

[0043] The Bayesian-based node reliability model can be expressed as: ; where, denotes the node normal rate (0-1), estimated from historical detection data; denotes the number of detection normal times, and needs to satisfy ; denotes the number of detection failure times; is the total number of detection times, denotes the normal rate threshold (e.g., 0.85). The minimum number is determined by the confidence threshold and the acceptable failure rate , i.e., , for example , , then .

[0044] The dynamic property health model can be expressed as: ; where, denotes the linearity error, used to reflect the static accuracy; denotes the time constant deviation, used to measure the response speed (ideally equal to 0); denotes the damping ratio, used to determine the dynamic stability (ideal value is between 0.6-0.7); , and are weights, which can be assigned according to the sensor type; and are decay coefficients, used to control the influence strength of parameter deviation on health.

[0045] The data-driven health model can be expressed as: ; where, denotes the anomaly index calculated based on a machine learning model (such as LSTM prediction error or Isolation Forest anomaly score); denotes the drift index, used to represent the long-term stability deviation; and are loss weights, and ; denotes the number of data points in the evaluation window.

[0046] Thus, the health degree level can be calculated and the level division and decision can be performed, as shown in Table 1.

[0047] Table 1 Health degree level division and decision

[0048] The sensor health degree evaluation model quantifies the health degree based on three dimensions of Bayesian reliability, dynamic response stability and data abnormality, and takes into account the historical performance and real-time state. The device health degree index is obtained through weighted calculation, realizing real-time monitoring and early warning of the device state. The sensors with a device health degree index lower than the set standard are adjusted or replaced to ensure that the collected data is always real and effective. Understandably, the set standard can be adjusted according to actual needs, and the present embodiment does not make specific limitations thereto.

[0049] In step S2, a mine noise fingerprint library is constructed, and the obtained real-time data is subjected to scene adaptive noise filtering.

[0050] The mine noise fingerprint library is constructed, and typical noise samples including blasting vibration, mechanical resonance and electromagnetic interference are collected, and the time-frequency feature codebook of the noise is extracted through wavelet packet decomposition technology. As an optional embodiment, the corresponding data can be collected through the laid sensors. During the collection process, the mine production operation log is recorded synchronously, and information such as the specific time of blasting operation, the start-stop period of mechanical equipment, the running state of the power supply system and the like is recorded, so as to accurately label the collected noise samples subsequently. For the three types of core noise of blasting vibration, mechanical resonance and electromagnetic interference, representative signal segments in different time periods are intercepted as samples, and the noise type, collection time and collection position corresponding to the samples are labeled. The wavelet packet decomposition technology is used to process the collected noise samples, and the noise signal is decomposed into components of different frequency bands through the decomposition process, and then the time-frequency features capable of representing the characteristics of the noise in each frequency band are extracted. After the extracted features are sorted and normalized, the correspondence between the noise type and the time-frequency features is established, and the time-frequency feature codebook is formed. The codebook and the labeling information of the samples (noise type, collection scene, etc.) are stored together to construct the mine noise fingerprint library, providing a feature basis for subsequent noise recognition and filtering.

[0051] The scene adaptive noise filtering is implemented, namely: 1) Based on the mining plan, strong filtering is performed during the blasting period. Specifically, first, according to the mining plan, the specific period of blasting operation, the operation area and the related operation parameters are determined, and the monitoring area range that needs to be strongly filtered is determined accordingly. In the blasting period, the real-time data collected in the area is processed using a filtering algorithm that can effectively suppress strong interference of blasting. The interference signals in the main frequency band of blasting vibration are first preliminarily removed, and then the residual high-frequency interference is further weakened through secondary filtering, so as to ensure that the blasting noise is filtered while the effective signals related to mine earthquake are preserved as much as possible. It should be noted that the filtering algorithm used only needs to be able to effectively suppress strong interference, and the present embodiment does not make specific limitations.

[0052] 2) The migration learning denoising model is used to suppress noise in non-blasting period data. Specifically, the denoising model is pre-trained using public mine non-blasting noise data sets and other mine non-blasting period noise data, so that the model has basic noise suppression capability. Subsequently, the noise data labeled in the current mine non-blasting period is selected to fine-tune the pre-trained model, so that the model adapts to the unique characteristics of the current mine non-blasting period noise. In the non-blasting period, the real-time data collected is input into the fine-tuned migration learning denoising model, and the noise in the data is suppressed through model operation, and the effective data after denoising is output.

[0053] 3) Quality classification control is performed on the filtered data according to the dynamic signal-to-noise ratio threshold. Specifically, the signal-to-noise ratio of the historical filtered data is analyzed, and the basic reference range of the signal-to-noise ratio of each period is determined according to the noise environment difference of different periods (such as daytime production period and night rest period), and the initial three-level quality thresholds of good, good and poor are set. After that, the signal-to-noise ratio statistical data of each period is updated once every fixed time, and the quality thresholds of each level are dynamically adjusted; if an abnormal working condition of the mine occurs (such as equipment failure and extreme weather), the threshold range is temporarily adjusted to ensure that the threshold can adapt to the current noise environment. The signal-to-noise ratio of the filtered data is calculated, and the data is divided into different quality levels according to the dynamic threshold. The optimal data is directly used for subsequent processing, the good data is used after simple optimization, the poor data needs to be further verified or removed, and the invalid data is directly discarded.

[0054] In step S3, the denoised data is classified into event types based on a dual-channel CNN architecture.

[0055] The dual-channel CNN architecture comprises a first channel, a second channel, a fusion layer, a fully connected layer, and a classifier. In the specific implementation process, the features of the first channel and the second channel are first spliced with physical features through the fusion layer, and then output the event type probability through the fully connected layer and the classifier. Among them, the first channel uses 1D-CNN to process the original waveform segment of the collected data, and the second channel uses 2D-CNN to process the time-frequency graph generated by the short-time Fourier transform. Specifically: 1) Select the effective data filtered in step S2, segment it into original waveform segments according to a fixed time length, normalize each segment, and unify the data range as the input of 1D-CNN. 1D-CNN extracts local features in the original waveform segment through multiple convolution operations, and then compresses and filters the features through the pooling operation to retain key feature information. Finally, the extracted features are integrated through the fully connected layer to output a feature vector that can represent the characteristics of the original waveform.

[0056] 2) For the same original waveform segment as the first channel, a short-time Fourier transform is used to convert it from a time-domain signal to a time-frequency matrix containing time and frequency information, and then the matrix is converted into a time-frequency graph in grayscale form. After adjusting the size of the time-frequency graph, it is used as the input of 2D-CNN. 2D-CNN extracts spatial features and frequency distribution features in the time-frequency graph layer by layer through multiple convolution and pooling operations. After multiple feature extraction and selection, the features are integrated through the fully connected layer to output a feature vector that can represent the characteristics of the time-frequency graph.

[0057] On this basis, the denoised data can be classified into event types based on the dual-channel CNN architecture. Specifically: 1) First, the original waveform feature vector output by the first channel is spliced with the time-frequency graph feature vector output by the second channel to form a fusion feature vector. To balance the weights of the features of the two channels, an attention mechanism is introduced to weight the fusion feature vector to highlight the feature information that is more valuable for event type classification. After weighting the fusion feature vector, the classification network is connected, the features are further processed through the fully connected layer, and finally the classifier outputs the probability values of each type of event (such as mine earthquake, blasting, noise, etc.) corresponding to the data.

[0058] 2) Collect historical event data in the mine, label the event type, and divide it into a training set, a validation set, and a test set for training and optimizing the dual-channel CNN model. During training, adjust the model parameters according to the classification accuracy of the model on the validation set. After the model is trained and the performance is up to standard, input the denoised data processed in step S2 into the model, and the model outputs the probability values of each type of event. Select the event type with the highest probability as the final classification result of the data. If the highest probability does not meet the set standard, it is marked as a suspected event and needs to be further confirmed in combination with other data.

[0059] In step S4, classification inversion is performed based on the event type classification results to optimize the source parameters.

[0060] When the event type is classified as a mining tremor event, moment tensor inversion is performed, i.e.: First, determine the level of the mine tremor event. In this embodiment, using... As a criterion for determining significant mine seismic activity, As a criterion for determining micro-seismic activity, As a criterion for determining extremely minor mineral seismic activity; among them, This indicates the volume wave magnitude. Subsequently, a graded inversion is performed based on the level of the seismic event. When the seismic event is classified as a significant seismic event, moment tensor inversion is used for localization; when it is classified as a minor seismic event, waveform cross-correlation and array beamforming techniques are used for localization; when it is classified as an extremely minor seismic event, the detection rate is improved by optimizing the array configuration. Specifically: 1) The mine-induced seismic event was classified as a significant mine-induced seismic event. The location was determined by moment tensor inversion positioning (i.e., inversion positioning based on travel time difference). The epicentral distance was calculated by using the travel time difference of P-waves and S-waves arriving at different stations. The location of the epicenter (latitude, longitude, and depth) was determined by the intersection of data from multiple stations.

[0061] 2) The seismic event was classified as a micro-seismic event. Positioning is achieved using waveform cross-correlation and array beamforming techniques. Waveform cross-correlation utilizes the waveform similarity of seismic events to achieve high-precision relative positioning by calculating the cross-correlation coefficient between template events and target events. ; in, This represents the cross-correlation coefficient between template events and target events in an earthquake event. The waveform parameters represent the template event. The waveform parameters represent the target event. As an optional implementation, when... If so, it is determined to be a valid related event.

[0062] Seismic events with high signal-to-noise ratios and recorded by multiple stations were selected, and waveform segments from 0.5 s before the P-wave to 4.0 s after the S-wave were extracted; sliding calculations were performed on the continuous waveform data. Value, obtained by the median of absolute deviation ( Set a threshold to detect microseismic events (e.g.) Based on cross-correlation delay Correct for travel time difference and invert the source location by combining the geometric distribution of stations.

[0063] Beamforming adjusts the delay and weighting of signals from each substation in the array to superimpose signals in a specific direction, thereby enhancing the signal-to-noise ratio and extracting azimuth information. The beam power expression is as follows: ; wherein, represents the beam power, represents the azimuth angle corresponding time delay; represents the pre-processed seismic signal collected by the th acquisition station (or sub-station in the station array), which varies with time; represents the time variable, which is used to describe the change process of the seismic signal in the time dimension, to ensure that the beam power calculation can reflect the superposition effect of the signal at different times.

[0064] Thereby, the instrument response can be corrected, the environmental noise can be removed by filtering (0.1-10 Hz), the azimuth angle and the elevation angle can be searched with a step size of 0.1° to locate the energy peak direction; when the epicentral distance is less than 500 m, the spherical wave model is used to correct the time delay to avoid the error of the plane wave assumption. First, the microseismic event is detected by cross-correlation, and then the beam forming is used for accurate positioning, which is suitable for hydraulic fracturing monitoring; combined with the waveform superposition and relative positioning idea, the influence of unknown excitation time of the source is eliminated to improve the calculation efficiency. A lightweight CNN model is embedded in the station node to extract the time-frequency features in real time and dynamically control the power consumption.

[0065] When the event type is divided into blasting or noise, the basic positioning inversion is performed, specifically: 1) Basic positioning inversion of blasting event The data determined as a blasting event in step S3 is extracted, and the time when the signal first arrives at each node is determined from the event data recorded by multiple acquisition nodes. Combined with the three-dimensional coordinates of each acquisition node in the mine geographic information system, one node is selected as a reference node, and the signal arrival time difference between other nodes and the reference node is calculated. Based on the propagation law of signals in rock, a positioning model is established, and the time difference and node coordinates are substituted into the model to solve the approximate coordinate position of the blasting source by calculation. The inversion obtained source coordinates are compared with the preset region of blasting operation in the mining plan to verify the accuracy of the positioning result, if the deviation is large, the data is rechecked and the inversion is performed again.

[0066] 2) Basic positioning inversion of noise event First, the data determined as noise events is matched with the feature codebook in the mine noise fingerprint library, combined with the signal characteristics of the noise, to determine the specific type of noise (such as mechanical resonance noise, electromagnetic interference noise). For different types of noise, corresponding positioning methods are used: for mechanical resonance noise, according to the noise signal intensity collected by multiple nodes, combined with the relationship between signal intensity and propagation distance, the distance from the noise source to each node is calculated, and then combined with the node coordinates to determine the noise source position through triangular positioning; for electromagnetic interference noise, by analyzing the phase difference of interference signals collected by different nodes, the incident direction of the interference signal is determined, and the intersection of the incident directions of multiple nodes is used to determine the interference source position. Compare the positioning results with the distribution of mine equipment and facilities, verify the positioning accuracy, and record the noise source information.

[0067] Fuse the event type classification result and the source parameter to generate a mine earthquake catalog, and make decisions and give warnings according to the corresponding response mechanism in the mine earthquake catalog. Specifically: 1) Mine earthquake catalog generation For the event types (mine earthquakes, blasting, etc.) classified in step S3, the corresponding source parameters are calculated, including the magnitude, source depth, and time of the event. Among them, the magnitude is calculated according to the signal amplitude recorded by the collection node combined with the relevant calibration data, the source depth is determined according to the source coordinates and mine geological data, and the time of the event is based on the signal arrival time recorded by the reference node. Integrate the event type classification result with the calculated source parameters, collection node information, data quality level, etc. into a unified format to build a mine earthquake catalog, which contains unique identification number, event core information, parameter calculation basis, etc. fields, and is stored in the database. At the same time, generate a regular report for relevant departments to review.

[0068] 2) Decision and warning based on the catalog According to the type and magnitude of the event in the mine earthquake catalog, four levels of warning are set: level I (red) warning corresponds to strong mine earthquake or strong blasting event that may cause serious disaster, level II (orange) warning corresponds to medium-intensity mine earthquake or blasting event that may cause local impact, level III (yellow) warning corresponds to weak mine earthquake or mechanical noise event that needs to be paid attention to, and level IV (blue) warning corresponds to weak mine earthquake or ordinary electromagnetic noise event that has little impact on production.

[0069] 3) Regular update of mine earthquake catalog.

[0070] If the warning level is triggered, an early warning information containing event details, warning level and suggested measures is generated immediately. According to the different warning levels, the warning information is pushed to the corresponding person in charge: level I and II warnings are pushed to mine senior management and front-line operation leaders, and level III and IV warnings are pushed to middle-level management and regional leaders.

[0071] Upon receiving the early warning, relevant personnel can execute preset emergency measures, for example, level I early warning requires immediate stop of operation and organization of personnel evacuation, level II early warning requires suspension of operation in the dangerous area and inspection of facilities, level III early warning requires arrangement of personnel to overhaul equipment or monitor mine-shaking areas, and level IV early warning requires recording of information and regular analysis. After the risk is eliminated or the problem is solved, the early warning is released according to the process, and the early warning processing condition is recorded to the system.

[0072] Embodiment two The embodiment discloses a mine earthquake catalog construction system based on double-drive classification inversion.

[0073] The mine earthquake catalog construction system based on double-drive classification inversion comprises: A multi-source sensor network module is configured to collect real-time data during mine vibration and mining activities, and perform energy detection; at the same time, the collection mode is automatically switched according to the energy detection result; A scene adaptive noise filtering module is configured to construct a mine noise fingerprint library, and perform scene adaptive noise filtering on the obtained real-time data; An event type division module is configured to divide the noise-removed data into event types based on a double-channel CNN architecture; A classification inversion module is configured to perform classification inversion according to the event type division result, to realize optimization of the seismic source parameters, specifically: when the event type is divided into a mine-shaking event, perform moment tensor inversion; when the event type is divided into blasting or noise, perform basic positioning inversion; and fuse the event type division result and the seismic source parameters to generate a mine earthquake catalog. Embodiment three The embodiment aims to provide a computer-readable storage medium.

[0074] The computer-readable storage medium stores a computer program, and the program is executed by a processor to implement the steps in the mine earthquake catalog construction method based on double-drive classification inversion according to the embodiment one of the present disclosure.

[0075] Embodiment four The embodiment aims to provide an electronic device.

[0076] The electronic device comprises a memory, a processor, and a program stored in the memory and executable on the processor, and the processor executes the program to implement the steps in the mine earthquake catalog construction method based on double-drive classification inversion according to the embodiment one of the present disclosure.

[0077] The steps involved in the apparatuses of the above embodiments two, three and four correspond to the method of embodiment one, and the specific implementation can refer to the relevant description of embodiment one. The term "computer readable storage medium" should be understood as including a single medium or multiple media of one or more instruction sets; it should also be understood as including any medium capable of storing, encoding or carrying the instruction set for execution by the processor and causing the processor to perform any of the methods in the present application.

[0078] Those skilled in the art should understand that each module or step of the present application described above can be realized by a general computer device, alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device for execution by a computing device, or they can be respectively made into each integrated circuit module, or a plurality of modules or steps among them can be made into a single integrated circuit module to realize. The present application is not limited to any specific combination of hardware and software.

[0079] Although the specific embodiments of the present application are described above in combination with the drawings, it is not a limitation on the scope of protection of the present application, and those skilled in the art should understand that various modifications or changes made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the scope of protection of the present application.

Claims

1. A mine seismic catalog construction method based on double drive classification inversion, characterized in that, The method comprises the following steps: Collecting real-time data during mine vibration and mining activities by using a multi-source sensor network, and performing energy detection; The multi-source sensor network automatically switches the collection mode according to the energy detection result; Constructing a mine noise fingerprint library and performing scene-adaptive noise filtering on the obtained real-time data; Dividing the denoised data into event types based on a double-channel CNN architecture; Performing classified inversion according to the event type division result to realize optimization of the seismic source parameters, specifically: when the event type is divided into a mine earthquake event, performing moment tensor inversion; when the event type is divided into blasting or noise, performing basic positioning inversion; and fusing the event type division result and the seismic source parameters to generate a mine earthquake catalog.

2. The mine seismic catalog construction method based on double drive classification inversion according to claim 1, characterized in that, The multi-source sensor network is a surface- underground three-dimensional monitoring network composed of surface wave sensors, wideband accelerometers and distributed optical fiber sensors.

3. The mine seismic catalog construction method based on double drive classification inversion according to claim 1, characterized in that, The multi-source sensor network automatically switches the collection mode according to the energy detection result, including: automatically switching to a down-sampling processing mode when a large energy event is detected; and automatically switching to an up-gain processing mode when a micro energy event is detected.

4. The mine seismic catalog construction method based on double drive classification inversion of claim 1, wherein, The scene-adaptive noise filtering comprises: performing strong filtering during the blasting period based on the mining plan, using a transfer learning denoising model to suppress noise in the non-blasting period, and controlling the quality classification of the filtered data according to a dynamic signal-to-noise ratio threshold.

5. The mine seismic catalog construction method based on double drive classification inversion according to claim 1, characterized in that, The double-channel CNN architecture comprises: a first channel, a second channel, a fusion layer, a full connection layer and a classifier; wherein the features of the first channel and the second channel are first spliced with physical features through the fusion layer, and then output the event type probability through the full connection layer and the classifier.

6. The mine seismic catalog construction method based on double drive classification inversion according to claim 5, characterized in that, The first channel uses 1D-CNN to process the original waveform segment of the collected data, and the second channel uses 2D-CNN to process the time-frequency graph generated by short-time Fourier transform.

7. The mine seismic catalog construction method based on double drive classification inversion of claim 1, wherein, When the event type is divided into a mine earthquake event, the moment tensor inversion is performed, including: determining the level of the mine earthquake event, when the level belongs to a significant mine earthquake, using moment tensor inversion positioning; when the level belongs to a micro mine earthquake, using waveform cross-correlation and array beamforming technology positioning; and when the level belongs to an extremely micro mine earthquake, improving the detection rate by optimizing the array configuration.

8. A mine seismic catalog construction system based on double drive classification inversion, characterized in that, The method comprises the following steps: The multi-source sensor network module is configured to collect real-time data during mine vibration and mining activities, and perform energy detection; at the same time, automatically switch the collection mode according to the energy detection result; The scene-adaptive noise filtering module is configured to construct a mine noise fingerprint library, and perform scene-adaptive noise filtering on the obtained real-time data; The event type division module is configured to divide the denoised data into event types based on a double-channel CNN architecture; The classified inversion module is configured to perform classified inversion according to the event type division result to realize optimization of the seismic source parameters, specifically: when the event type is divided into a mine earthquake event, performing moment tensor inversion; when the event type is divided into blasting or noise, performing basic positioning inversion; and fusing the event type division result and the seismic source parameters to generate a mine earthquake catalog.

9. A computer-readable storage medium having stored thereon a program, characterized in that, The program is executed by the processor to implement the steps in the method for constructing a mine seismic catalog based on double-drive classification inversion according to any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and capable of running on the processor, characterized by The processor executes the program to implement the steps in the method for constructing a mine seismic catalog based on double-drive classification inversion according to any one of claims 1-7.

Images