Deep coal mine disaster early warning method and system based on geological radar

Abstract

The invention discloses a deep coal mine disaster early warning method and system based on a geological radar, and relates to the technical field of disaster early warning, and the method comprises the steps: carrying out the monitoring coverage analysis of a target monitoring region, laying geological radar equipment according to a result, and constructing a geological radar array; scanning a mine underground structure through an array to obtain a rock stratum data array; performing spatial hierarchical feature recognition on the rock stratum data array to obtain rock stratum risk feature distribution; calling historical data to obtain a plurality of historical risk feature distributions; and performing risk trend prediction on the rock stratum risk feature distribution according to the historical distribution, and outputting a predicted disaster critical working condition. The technical problem that the real-time performance and the accuracy of disaster risk identification are insufficient due to the fact that a traditional coal mine disaster early warning method depends on low-frequency and single-dimension monitoring data is solved, and the technical effect of improving the accuracy and the real-time performance of coal mine disaster early warning is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of disaster warning, and particularly to a deep coal mine disaster warning method and system based on ground penetrating radar. Background Art

[0002] With the continuous increase of coal mining depth, the geological disaster risks faced by deep coal mines are becoming increasingly severe. Due to relying on low-frequency and single-dimensional data collection, traditional disaster monitoring methods are difficult to meet the requirements of real-time, accuracy, and comprehensiveness. The geological conditions of deep coal mines are complex, and phenomena such as rock layer deformation, crack expansion, and stress concentration occur frequently, which are extremely likely to cause disasters such as roof falls, collapses, and gas outbursts, seriously threatening the lives of miners and the safety of mine production. Existing monitoring technologies are mostly based on point sensors or manual inspections, with problems such as limited coverage, lagging data updates, and insufficient warning accuracy, and cannot achieve all-round and high-frequency monitoring of the geological structure of deep coal mines. In addition, traditional methods lack the ability of intelligent identification and dynamic prediction of rock layer risk characteristics, and it is difficult to provide effective warning information before disasters occur. Therefore, there is an urgent need for an intelligent warning method and system that can monitor the changes in the geological structure of deep coal mines in real time, accurately identify potential risk areas, and dynamically predict the probability of disasters, so as to improve the safety level of coal mine production and ensure the lives and property safety of miners. Summary of the Invention

[0003] The present application provides a deep coal mine disaster warning method and system based on ground penetrating radar, which is used to solve the technical problem that the real-time and accuracy of disaster risk identification are insufficient due to the traditional coal mine disaster warning method relying on low-frequency and single-dimensional monitoring data, and achieve the technical effect of accurately analyzing the rock layer risk characteristics by high-frequency ground penetrating radar real-time scanning and multi-level spatial feature identification, combined with deep learning algorithms, and improving the accuracy and real-time of coal mine disaster warning.

[0004] In view of the above problems, the present application provides a deep coal mine disaster warning method and system based on ground penetrating radar.

[0005] In the first aspect of the present application, a deep coal mine disaster warning method based on ground penetrating radar is provided. The method includes: performing monitoring coverage analysis on a target monitoring area, and arranging ground penetrating radar devices according to the analysis results to complete the construction of a ground penetrating radar array; performing underground structure scanning of the mine on the target monitoring area through the ground penetrating radar array to obtain a rock layer data array; obtaining a rock layer risk feature distribution by performing spatial hierarchical feature identification on the rock layer data array; obtaining a plurality of historical risk feature distributions by calling historical data; and performing risk trend prediction on the rock layer risk feature distribution based on the plurality of historical risk feature distributions, and outputting a predicted disaster critical condition, where the predicted disaster critical condition has a predicted disaster time identifier.

[0006] In the second aspect of the present application, a deep coal mine disaster early warning system based on ground penetrating radar is provided. The system includes: An equipment layout component: Analyze the monitoring coverage of the target monitoring area, and layout the ground penetrating radar equipment according to the analysis results to complete the construction of the ground penetrating radar array; An underground structure scanning component: Perform underground structure scanning of the target monitoring area through the ground penetrating radar array to obtain a rock formation data array; A feature recognition component: Identify the spatial hierarchical features of the rock formation data array to obtain the distribution of rock formation risk features; A historical data calling component: Obtain multiple historical risk feature distributions through historical data calling; A risk trend prediction component: Perform risk trend prediction on the distribution of rock formation risk features based on the multiple historical risk feature distributions, and output a predicted disaster critical condition, where the predicted disaster critical condition has a predicted disaster time identifier.

[0007] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0008] Analyze the monitoring coverage of the target monitoring area, and layout the ground penetrating radar equipment according to the analysis results to complete the construction of the ground penetrating radar array; Perform underground structure scanning of the target monitoring area through the ground penetrating radar array to obtain a rock formation data array; Identify the spatial hierarchical features of the rock formation data array to obtain the distribution of rock formation risk features; Obtain multiple historical risk feature distributions through historical data calling; Perform risk trend prediction on the distribution of rock formation risk features based on the multiple historical risk feature distributions, and output a predicted disaster critical condition, where the predicted disaster critical condition has a predicted disaster time identifier. Achieve the technical effect of accurately analyzing the rock formation risk features through high-frequency ground penetrating radar real-time scanning and multi-level spatial feature recognition, and combining deep learning algorithms to improve the accuracy and real-time performance of coal mine disaster early warning. Description of the Drawings

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0010] Figure 1 It is a schematic flow chart of a deep coal mine disaster early warning method based on ground penetrating radar provided by an embodiment of the present application;

[0011] Figure 2 It is a schematic structural diagram of a deep coal mine disaster early warning system based on ground penetrating radar provided by an embodiment of the present application.

[0012] Description of the attached drawing reference numerals: Equipment layout component 11, underground structure scanning component 12, feature recognition component 13, historical data calling component 14, risk trend prediction component 15. Detailed implementation manners

[0013] This application provides a deep coal mine disaster warning method and system based on ground penetrating radar, which is used to solve the technical problem that the traditional coal mine disaster warning method relies on low-frequency and single-dimensional monitoring data, resulting in insufficient real-time performance and accuracy of disaster risk identification, and achieves the technical effect of improving the accuracy and real-time performance of coal mine disaster warning through high-frequency ground penetrating radar real-time scanning and multi-level spatial feature recognition, combined with deep learning algorithms to accurately analyze the risk characteristics of rock formations.

[0014] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts shall fall within the protection scope of this application.

[0015] Embodiment 1, as Figure 1 shown, this application provides a deep coal mine disaster warning method based on ground penetrating radar, and the method includes:

[0016] Conduct a monitoring coverage analysis on the target monitoring area, and arrange ground penetrating radar equipment according to the analysis results to complete the construction of the ground penetrating radar array.

[0017] In one embodiment, in a coal mine disaster warning system, reasonably arranging geological radar equipment is a key step to ensure the monitoring effect. First, it is necessary to conduct a monitoring coverage analysis of the target monitoring area, that is, to evaluate the monitoring requirements of different areas according to factors such as the geological structure of the mine, the mining depth, and the rock stratum structure. This analysis mainly includes multiple parameters such as the monitoring accuracy requirements, the equipment resolution, and the monitoring coverage. For high-risk disaster areas, for example, areas where rock deformation has occurred or there is an accumulation of historical fissures, a higher monitoring density should be set to improve the detection accuracy. After completing the monitoring coverage analysis, it is necessary to optimize the layout plan of the geological radar equipment. This process usually uses a layout site optimization algorithm to calculate and screen out the optimal equipment layout points based on constraints such as monitoring accuracy and coverage, and finally form a set of layout site arrays. After completing the optimization of the equipment layout points, it is necessary to conduct batch debugging of the geological radar equipment to ensure that the equipment can accurately obtain rock stratum information during actual monitoring. After the debugging is completed, according to the optimized layout plan, the geological radar equipment is gradually installed in the mine and connected and debugged with the data acquisition system, and finally the construction of the geological radar array is completed. The formation of this array enables the system terminal to scan the underground structure of the mine in all directions and from multiple angles, thereby providing high-precision data support for subsequent rock stratum risk identification and disaster prediction.

[0018] In a possible implementation manner, a monitoring coverage analysis is performed on the target monitoring area, and the geological radar equipment is arranged according to the analysis result to complete the construction of the geological radar array. The method includes:

[0019] Collect monitoring correlation features of the target monitoring area to determine monitoring performance indicators, where the monitoring performance indicators include target monitoring accuracy, equipment resolution, and monitoring coverage; optimize the layout sites with the target monitoring accuracy and monitoring coverage as constraints to obtain a layout site array; after batch debugging of the geological radar equipment according to the target monitoring accuracy, arrange the geological radar equipment with reference to the layout site array to complete the construction of the geological radar array.

[0020] Optionally, during the construction of the ground penetrating radar array, to ensure the monitoring accuracy and coverage, it is first necessary to collect detailed monitoring correlation features of the target monitoring area to clarify appropriate monitoring performance indicators. Specifically, first, through geological exploration data, rock formation characteristics, mining depth, and historical disaster data, analyze the rock formation distribution, main fault zones, and potential dangerous areas inside the mine, determine the monitoring requirements for different areas. For example, areas with high stress concentration, historical collapse areas, etc. require higher monitoring accuracy, while relatively stable areas can appropriately reduce the monitoring density to optimize the equipment layout resources. Combining these analysis results, determine the target monitoring accuracy for each area, and calculate its optimal monitoring coverage in combination with the equipment resolution of the ground penetrating radar. After clarifying the monitoring performance indicators, it is necessary to optimize the layout sites, that is, find the equipment layout positions that can not only meet the monitoring accuracy requirements but also achieve the maximum coverage. In this process, the target monitoring area will be gridified through digital modeling and optimization algorithms, and the initial monitoring points will be located in the grids. Then, based on the monitoring contribution evaluation, the monitoring points will be gradually screened and adjusted until the optimal layout site array is obtained. After completing the optimization of the layout sites, batch debugging of the ground penetrating radar equipment is carried out to ensure that all equipment can work normally and meet the preset monitoring accuracy standards. The specific debugging process includes the following: First, adjust the signal transmission power, that is, according to the penetration ability of the radar in different media, adjust the transmission power to ensure that the signal can effectively penetrate the rock formation while avoiding signal interference caused by too high power; Second, calibrate the receiving sensitivity, that is, optimize the sensitivity of the radar receiving module so that it can accurately receive the reflected echoes, reduce noise interference, and improve the signal analysis ability; Third, optimize the data acquisition rate, that is, combine the dynamic change characteristics of the mine environment to adjust the data acquisition frequency to ensure that the changes in the rock formation can be reflected in real time. After the debugging is completed, referring to the optimized layout site array, the ground penetrating radar equipment is gradually installed in the mine according to the preset plan, and overall joint debugging is carried out to ensure that all equipment can stably collect data and synchronously transmit information. Finally, the construction of the ground penetrating radar array is completed, enabling the system terminal to perform high-precision and all-round real-time scanning of the underground structure of the mine, providing reliable data support for subsequent disaster warnings.

[0021] In a possible implementation manner, with the target monitoring accuracy and monitoring coverage as constraints, optimize the layout sites to obtain a layout site array. The method includes:

[0022] Digitally process the target monitoring area to obtain a three-dimensional geological model; preset a grid scale, and grid the three-dimensional geological model according to the grid scale to obtain a three-dimensional grid unit model; configure an initial monitoring spacing according to the target monitoring accuracy; locate K initial monitoring grid points in the three-dimensional grid unit model according to the initial monitoring spacing; taking the target monitoring accuracy and monitoring coverage as constraint conditions, conduct a monitoring contribution evaluation on the K initial monitoring grid points, and screen and update the grid point positions according to the evaluation results until the layout site array is output.

[0023] Optionally, when optimizing the layout sites, it is necessary to ensure the reasonable distribution of radar equipment through digital modeling and layout site optimization to achieve high-precision and large-scale mine monitoring. Specifically, first, the target monitoring area needs to be digitized. That is, based on information such as mine surveying data and geological exploration data (such as drilling records, seismic wave reflection data), a high-precision three-dimensional geological model is constructed. This model usually uses a three-dimensional coordinate system to represent the spatial structure of the mine, including key geological information such as roadway layout, rock layer distribution, and fault structure. The data sources can include mine surveying data (such as laser scanning or photogrammetry), historical geological section data, seismic wave reflection imaging, borehole exploration data, etc. By using computer modeling techniques such as BIM (Building Information Modeling), GIS (Geographic Information System), and finite element analysis software to fuse these data, the required three-dimensional digital model of the mine, that is, the three-dimensional geological model, can be formed as the basis for subsequent optimization of the layout of monitoring points. To facilitate the analysis and optimization of the layout positions of radar equipment, it is necessary to perform grid processing on the three-dimensional geological model. In this process, the grid scale will be set, that is, the spatial size of each grid unit is defined, usually determined according to factors such as the mine exploitation scale and the detection accuracy of geological radar. For example, for high-risk areas, a smaller grid scale (such as 1m×1m×1m) can be used, while for stable areas, a larger scale (such as 5m×5m×5m) can be used. Then, using regular grid or irregular grid division methods, the three-dimensional model of the mine is discretized according to the preset grid scale to form a three-dimensional model of grid units. Among them, each grid unit represents an independent monitoring area of the mine. Through grid processing, the three-dimensional geological model is divided into several small unit grids, providing data support for subsequent layout of monitoring points. Subsequently, the initial monitoring interval is configured according to the target monitoring accuracy. Among them, the target monitoring accuracy determines the layout density of geological radar. If the monitoring accuracy requirement is high (such as detecting microfractures and local stress concentration areas), the monitoring interval needs to be reduced, that is, the equipment density is increased. If the monitoring area is large and mainly focuses on overall structural changes (such as large-scale rock layer deformation), the monitoring interval can be expanded, the number of radar equipment can be reduced, and the coverage efficiency can be improved. According to these requirements, the required initial monitoring interval, that is, the minimum layout distance between equipment, can be determined. For example, in high-precision areas (such as faults and goafs), the interval is set to 1m - 2m; in medium-precision areas (such as rock layer continuity monitoring), the interval is set to 3m - 5m; in low-precision areas (such as stable rock layers far from the excavation face), the interval is set to 5m - 10m.After that, in the 3D model of the grid cell, candidate monitoring grid points are evenly generated along the three directions of X, Y, and Z according to the initial monitoring spacing. For example, a point is generated at a certain interval (such as 5 meters) on the horizontal plane, and at the same time, stratified monitoring points are generated in the vertical direction according to the distribution characteristics of the rock strata. Then, the generated candidate monitoring grid points are screened to eliminate invalid points. Invalid points include: points located in non-monitoring areas (such as mined areas or irrelevant geological areas); points located in areas where the terrain or geological conditions are not suitable for equipment installation (such as steep slopes or areas with too high groundwater levels); redundant points that overlap with other points too much. According to the screened candidate monitoring grid points, K points are selected as the initial monitoring grid points. The selection of these initial points follows any one of the following principles, such as uniform distribution (covering the entire target area as much as possible), prioritizing high-risk areas (preferably arranging in fault zones and stress concentration areas to increase the point density), and avoiding signal interference (avoiding too much signal overlap between multiple radar devices, which affects the detection accuracy). Specifically, it can be determined according to the actual business needs. After the optimization is completed, the determined K initial monitoring grid points are recorded as the initial layout plan, which serves as the basis for subsequent optimization and adjustment. These points will be used for the preliminary layout of the ground penetrating radar equipment and provide references for the subsequent monitoring contribution evaluation and point optimization. To optimize the layout of the monitoring points, the contribution degree of the K initial monitoring grid points is evaluated, that is, the contribution degree of each monitoring point to the overall monitoring effect is analyzed. The evaluation indicators include coverage contribution, accuracy contribution, and redundancy. Among them, the coverage contribution refers to the range of the monitoring area that the point can cover; the accuracy contribution refers to the degree of improvement of the monitoring accuracy by the point; the redundancy refers to the coverage overlap situation between the point and other points. For the coverage contribution, it can be calculated by taking the ratio of the coverage area of the point (calculated by the effective detection radius of the monitoring equipment) to the total area of the target monitoring area; for the accuracy contribution, it is calculated by taking the reciprocal of the distance from the point to its nearest neighbor point. The smaller the distance, the higher the point density and the greater the accuracy contribution; for the redundancy, it is obtained by calculating the ratio of each coverage overlap area of the point to the coverage area of the point and then accumulating these ratios. The higher the redundancy, the lower the necessity of the point. Then, subtract the redundancy of the point from 1, and then perform a weighted calculation on the obtained difference with the coverage contribution and accuracy contribution of the point to obtain the comprehensive contribution value of the point. According to this comprehensive contribution value, high-contribution points are screened out, that is, points greater than the set threshold are retained, and points lower than the set threshold are eliminated. After the screening is completed, it is necessary to clarify which areas are still not fully covered, that is, calculate the union of the coverage areas of all the retained points to obtain the total coverage area, and then calculate the difference between the total area of the target monitoring area and the total coverage area to obtain the uncovered area. Take the center point of the uncovered area as the candidate location for the new points, and evenly arrange new points in the uncovered area according to the initial monitoring spacing.After the addition is completed, repeat the above calculation and screening process until the comprehensive contribution value of the target monitoring area meets the threshold requirements, and the coverage rate (the ratio of the total coverage area to the total area of the target monitoring area) and the monitoring accuracy (the average distance between the remaining points) also meet the preset constraint conditions. Through multiple rounds of iterative calculations, an optimized layout site array is finally obtained to ensure that the layout of the ground penetrating radar equipment can not only meet the monitoring accuracy requirements but also achieve the maximum coverage range. In summary, through the above process, the scientific design from the initial monitoring grid points to the optimized layout site array is realized, providing a reliable basis for the efficient layout of the ground penetrating radar equipment and the early warning of deep coal mine disasters, ensuring more accurate, real-time, and efficient monitoring of the underground structure of the mine.

[0024] Perform underground structure scanning of the target monitoring area through the ground penetrating radar array to obtain a rock formation data array.

[0025] In one embodiment, the installed ground penetrating radar equipment is networked and started to ensure that all equipment works synchronously. Each device emits and receives signals according to preset parameters (such as transmission frequency, sampling rate, etc.). Subsequently, the ground penetrating radar equipment emits electromagnetic wave signals underground. When the signals encounter different rock formations or geological interfaces, they are reflected. The equipment receives the reflected signals and records the data. The scanning range covers the horizontal and vertical directions of the target monitoring area to ensure all-round detection. After that, each ground penetrating radar equipment converts the collected reflected signals into rock formation data, including information such as signal intensity, reflection time, waveform characteristics, and signal attenuation. These data reflect the characteristics of the underground rock formation structure, density, fracture distribution, etc. Then, the data collected by multiple ground penetrating radar equipment are integrated to form a three-dimensional rock formation data array. This array stores data in a grid form, and each grid unit corresponds to a spatial position and contains the rock formation characteristic information of that position. After obtaining the rock formation characteristic information, preprocessing of the rock formation data array is also carried out, including operations such as denoising, filtering, normalization, and signal enhancement to improve the data quality and provide a reliable basis for subsequent analysis. Through the above process, the ground penetrating radar array completes the underground structure scanning of the target monitoring area and generates a high-precision rock formation data array, providing data support for subsequent risk characteristic identification and disaster early warning.

[0026] Obtain the rock formation risk characteristic distribution by identifying the spatial hierarchical characteristics of the rock formation data array.

[0027] In one embodiment, after obtaining the rock formation data array, it is necessary to identify the spatial hierarchical features of the data to extract the key information of the underground structure and finally form the rock formation risk feature distribution. First, it is necessary to load the pre-constructed rock formation defect identification model, which can accurately identify the abnormal features in the rock formation data. After obtaining this model, the rock formation data array is loaded into the model, and key parameters such as the reflection signal intensity, waveform characteristics, and signal attenuation rate in the data are analyzed layer by layer to identify potential defects in the rock formation, such as fractures, faults, or stress concentration areas. These defect data are stored as the rock formation risk defect array. Subsequently, according to the coverage cross-relationship of the ground penetrating radar array, multiple rock formation risk defect arrays are spliced and integrated to form a complete rock formation risk feature distribution map, which can clearly display the risk features and their severity at different positions in the target monitoring area, providing a scientific basis for subsequent risk trend prediction and disaster warning. Through the spatial hierarchical feature identification and defect integration in the whole process, a comprehensive assessment of the rock formation risk is achieved.

[0028] In a possible implementation manner, by performing spatial hierarchical feature identification on the rock formation data array, a rock formation risk feature distribution is obtained, and the method includes:

[0029] Pre-construct a rock formation defect identification model; perform defect identification by loading the rock formation data array into the rock formation defect identification model to obtain a rock formation risk defect array; splice the rock formation risk defect arrays according to the coverage cross-relationship of the ground penetrating radar array to obtain the rock formation risk feature distribution.

[0030] Optionally, before performing the analysis of the rock formation risk characteristics, it is necessary to pre-construct a rock formation defect identification model. The core task of this model is to automatically identify possible defect areas from the rock formation data obtained by the ground penetrating radar, such as fissures, faults, stress concentration areas, and abnormal water-bearing areas. To this end, first, collect sample data of various rock formation defect types, including fissures, faults, stress concentration areas, etc., and perform defect feature annotation on these sample data to form a training data set. Then, use these annotated data to train multiple rock formation defect identification sub-models, each of which focuses on identifying a specific type of defect. Subsequently, according to the occurrence frequency and importance of different types of defects in practice, assign corresponding calculation weights to each sub-model, and integrate these sub-models in parallel to construct a comprehensive rock formation defect identification model, which can efficiently identify multiple defect types in the rock formation data. After that, extract data blocks of local areas from the rock formation data array and input them into the rock formation defect identification model. Each sub-model in the model will analyze the signal characteristics in the data block in parallel to identify potential defect types. The identification results of each sub-model will be comprehensively evaluated to generate the defect distribution of this area. Repeat this process until the entire rock formation data array is analyzed block by block, and finally generate a rock formation risk defect array containing all defect information. Then, since the ground penetrating radar array usually consists of multiple radar devices, there may be a certain overlapping and intersecting area in the detection ranges of different radar devices. Therefore, after obtaining the rock formation risk defect array, it is also necessary to perform splicing processing on the data. In this process, the coordinate transformation method will be used to project the data of each radar device onto the same geographic coordinate system to ensure that the data of different devices can be directly compared. Then, align and splice the data in the rock formation risk defect array according to the device coverage range to ensure that the defect data identified by different devices can be seamlessly connected. For the data in the overlapping area, weighted average or priority rules are used for fusion to eliminate redundancy and improve data accuracy. Finally, integrate the defect data of all devices into a complete three-dimensional rock formation risk characteristic distribution map, clearly showing the location, type, etc. of potential risks in the target monitoring area. This distribution map provides reliable data support for subsequent risk trend prediction and disaster warning.

[0031] In a possible implementation manner, a rock formation defect identification model is pre-constructed, and the method includes:

[0032] Interactively obtain multiple sample rock formation data of multiple types of rock formation defects; perform defect feature identification on the multiple sample rock formation data to obtain multiple sample defect identification data; use the multiple sample rock formation data and the multiple sample defect identification data as training data to construct multiple rock formation defect recognition sub-models; interactively obtain the multiple defect accident frequencies of the multiple types of rock formation defects; complete the construction of the rock formation defect recognition model by paralleling the multiple rock formation defect recognition sub-models and performing operation computing power allocation according to the multiple defect accident frequencies.

[0033] Optionally, first, through interaction with historical mine data, laboratory simulation data, etc., sample data of various types of rock formation defects, including cracks, faults, stress concentration areas, etc., are obtained. These sample data cover typical defect characteristics under different geological conditions and mining environments. Subsequently, these sample data are subjected to detailed defect feature identification, and the type of each defect (such as cracks, faults, stress concentration areas, water inrush risks, etc.) is marked to form multiple sample defect identification data. The marked sample rock formation data and sample defect identification data are then used as training data to train multiple rock formation defect identification sub-models, each of which focuses on identifying a specific type of defect, for example, one sub-model specifically identifies cracks, and another sub-model specifically identifies faults. Through machine learning algorithms (such as convolutional neural networks or support vector machines), each sub-model can learn the characteristic patterns of the corresponding defects and have independent defect recognition capabilities. In the process of constructing multiple rock formation defect recognition sub-models, a 3D convolutional neural network (3D CNN) is used to construct an initial model structure for each type of rock formation defect, and then the corresponding data is used for training. Taking the crack recognition sub-model as an example, a crack recognition sub-model is constructed using 3D CNN, including input layer, 3D convolution layer, pooling layer, fully connected layer, output layer, etc. The input layer is used to receive the three-dimensional rock formation data array, the 3D convolution layer is used to extract the spatial features of the data, the pooling layer is used to reduce the data dimension, the fully connected layer is used to integrate the features, and the output layer is used to generate the crack recognition results. After that, the model parameters are initialized, and the Kaiming normal distribution (HeInitialization) is used to assign weights to ensure that the gradient of the neural network is stable during the deep propagation process. The training set (consisting of sample rock formation data with the rock formation defect type of crack) is input into the initialized crack recognition sub-model for forward propagation. The data is subjected to 3D convolution layer, ReLU activation, pooling layer, fully connected layer, etc. for layer-by-layer feature learning, and the location and probability distribution of possible cracks in the rock formation are calculated. During the model training process, the cross-entropy loss function is used to calculate the error between the predicted results and the actual crack annotation data, and the gradient of the loss to the weight of each layer is calculated layer by layer through backpropagation. The Adam adaptive optimizer is used for parameter optimization, and the weights are continuously adjusted to minimize the loss function and improve the accuracy of crack detection. During the model training process, as the number of iterations increases, the model is gradually optimized so that it can accurately identify the crack area in the radar data. The above process is repeated until the maximum number of iterations is reached or the model converges to the set threshold.After the training is completed, the validation set is used to test the model and evaluate the accuracy of the model in the crack identification task. If the prediction accuracy of the model reaches the set standard, the current crack identification sub-model is output as the final model; otherwise, hyperparameters such as the learning rate, training batch, and regularization parameter are adjusted to further optimize the model to improve the stability and generalization ability of crack identification. After completing the training of the sub-model, the occurrence frequency data of various rock stratum defect types in practice are obtained through the statistics of historical data. These frequency data reflect the threat degree of different defect types to mine safety. Based on these frequency data, the corresponding operating computing power is allocated to each rock stratum defect identification sub-model. For example, the sub-model corresponding to the defect type with a higher occurrence frequency (such as cracks) will be allocated more computing resources to ensure its identification efficiency and accuracy. Finally, multiple rock stratum defect identification sub-models are integrated in parallel to construct a comprehensive rock stratum defect identification model. This model can simultaneously process the identification tasks of multiple defect types and dynamically allocate computing resources according to the actual threat degree of the defects, so as to achieve efficient and accurate analysis of rock stratum data. The construction of this model provides strong technical support for subsequent rock stratum risk feature identification and disaster warning.

[0034] In a possible implementation manner, by loading the rock stratum data array into the rock stratum defect identification model for defect identification to obtain a rock stratum risk defect array, the method includes:

[0035] Calling the first local rock stratum data from the rock stratum data array; after inputting the first local rock stratum data into the rock stratum defect identification model, performing parallel defect identification through multiple rock stratum defect identification sub-models in the rock stratum defect identification model to obtain multiple local rock stratum defect spaces; performing feature fusion on the multiple local rock stratum defect spaces to obtain the first rock stratum risk defect; and so on. After loading the rock stratum data array into the rock stratum defect identification model for defect identification, storing the identification results based on the array structure to obtain the rock stratum risk defect array.

[0036] Optionally, after the construction of the rock formation defect recognition model, it is necessary to identify the actually collected rock formation data array to detect potential rock formation defects such as fractures, faults, stress concentration areas, water inrush anomalies, etc. First, according to the divided grid, the first local data block is obtained from the rock formation data array as the first local rock formation data, which contains information such as the rock formation signal intensity and waveform characteristics in this local area. Subsequently, the first local rock formation data is input into the rock formation defect recognition model, and multiple rock formation defect recognition sub-models in the model will process this data block in parallel. Each sub-model focuses on identifying a specific type of defect (such as fractures, faults, stress concentration areas, etc.). By analyzing the signal characteristics in the data block, a defect distribution map within this local range is generated. The recognition result of each sub-model will be stored as a local rock formation defect space, which contains information such as the type and location of the defect. After that, the local rock formation defect spaces generated by each sub-model are spatially aligned to ensure that the data of different defect types are consistent in spatial position. Then, weighted fusion or priority rules are used to integrate the recognition results of different defect types into a comprehensive defect distribution. For example, for the same location, if multiple sub-models identify different types of defects, they are prioritized according to the severity or occurrence frequency of the defects, and the most severe defect type is retained. Finally, the first rock formation risk defect is generated, and this result comprehensively reflects all potential risk characteristics within this local range. And so on, each local data block in the rock formation data array is sequentially loaded into the rock formation defect recognition model for defect recognition and feature fusion. The processing result of each local data block will be stored according to its spatial position in the rock formation data array, and finally a complete three-dimensional rock formation risk defect array is formed. This array is stored in a grid structure, and each grid cell contains information such as the defect type and location at this position, providing comprehensive data support for the subsequent construction of the rock formation risk feature distribution and disaster warning.

[0037] Multiple historical risk feature distributions are obtained by calling historical data.

[0038] In one embodiment, a connection is established with the historical monitoring database. By calling the data in the historical monitoring database, multiple historical risk feature distributions are obtained. These historical risk feature distributions not only show the spatio-temporal evolution law of the rock formation risk but also provide an important reference basis for the subsequent risk trend prediction. By comparing and analyzing these historical distributions, the formation mechanism and development trend of the rock formation risk can be better understood, thereby improving the accuracy and reliability of disaster warning.

[0039] Based on the multiple historical risk feature distributions, risk trend prediction is performed on the rock formation risk feature distribution, and a predicted disaster critical condition is output, where the predicted disaster critical condition has a predicted disaster time identifier.

[0040] In one embodiment, a comparative analysis is performed on multiple historical risk feature distributions and the current rock formation risk feature distribution to identify the evolution law of the rock formation risk features. Then, based on these evolution laws, the changes in the rock formation risk features over a period of time in the future are predicted. For example, the expansion rate of fractures is predicted. Subsequently, the evolution time is calculated according to the prediction results to determine the time point when the critical condition is reached. After that, the time point when this critical condition is reached and the change rate are combined for defect evolution fitting to obtain the predicted disaster critical working condition. This result not only provides the possibility and time range of the disaster occurrence, but also provides a scientific decision-making basis for mine safety management, helping miners take preventive measures in time to avoid the occurrence of disasters.

[0041] In a possible implementation manner, based on the multiple historical risk feature distributions, a risk trend prediction is performed on the rock formation risk feature distribution, and the predicted disaster critical working condition is output. Then, the method further includes:

[0042] Decompose the predicted disaster critical working condition to obtain multiple disaster warning index thresholds; perform spatial hierarchical feature recognition on the monitoring data array transmitted back in real time by the ground penetrating radar array to obtain the real-time risk feature distribution; traverse the real-time risk feature distribution with the multiple disaster warning index thresholds; when any of the real-time risk feature distributions meets the multiple disaster warning index thresholds, output the associated audible and visual warning according to the index response type.

[0043] Optionally, after calculating the predicted critical conditions of disasters, it is necessary to decompose them to extract multiple disaster warning index thresholds, and combine with the monitoring data real-time transmitted by the ground penetrating radar array to analyze the real-time risk characteristic distribution, ensuring that when the detected risk exceeds the set threshold, an audible and visual warning can be triggered in a timely manner to remind miners to take safety measures. Specifically, in the prediction of critical disaster conditions, the possible time range and key triggering conditions of disasters have been calculated. To improve the accuracy of early warning, it is necessary to further decompose the critical conditions to extract multiple disaster warning index thresholds for real-time monitoring and comparison. The disaster warning indexes are mainly set based on the following factors: crack propagation rate, crack propagation length, stress concentration, etc. Each index corresponds to a threshold, and these thresholds are determined according to historical data, theoretical models and actual engineering experience, and are used to judge the critical conditions for disasters to occur. Subsequently, the real-time collected rock formation data is loaded into the rock formation defect identification model to identify the risk characteristics in the current rock formation, such as cracks, faults or stress concentration areas. Then, these risk characteristics are integrated according to their spatial positions to generate a real-time risk characteristic distribution, which reflects the risk state of the rock formation in the current monitoring area. After that, each risk characteristic in the real-time risk characteristic distribution is compared with the corresponding warning index threshold. For example, check whether the crack propagation length exceeds the threshold, and whether the range of the stress concentration area exceeds the safety limit. By traversing the entire distribution, the system terminal will record all situations that meet or exceed the threshold. When any risk characteristic in the real-time risk characteristic distribution meets multiple disaster warning index thresholds, an associated audible and visual warning will be output according to the index response type. For example, if the crack propagation length exceeds the threshold, a crack warning will be triggered; if the range of the stress concentration area exceeds the safety limit, a stress warning will be triggered. The warning information is transmitted to the miners in real time through the audible and visual alarm device, and at the same time, combined with the intelligent guidance system, a safe evacuation route is provided to ensure that the miners can take shelter in a timely manner. This process realizes the real-time monitoring and rapid response to disaster risks, effectively ensuring the safety production of the mine.

[0044] In a possible implementation manner, when any of the real-time risk characteristic distributions meets the multiple disaster warning index thresholds, an associated audible and visual warning is output according to the index response type. The method includes:

[0045] Sending the associated audible and visual warning to the intelligent guidance module to semi-automatically activate the intelligent guidance module; the intelligent guidance module starts an associated guidance indication based on the associated audible and visual warning.

[0046] Optionally, when it is detected that the real-time risk feature distribution meets the disaster warning index threshold, an associated audible and visual warning signal will be immediately generated. This signal is sent to the intelligent guidance module through the communication network to trigger the semi-automatic activation of the module. During the activation process, the environmental status of the current mine (such as roadway layout, equipment operation, etc.) will be checked to ensure that the intelligent guidance module is in an available state. The intelligent guidance module matches the corresponding associated guidance instructions according to the risk type in the warning signal (such as fissure, fault or stress concentration), and activates different evacuation guidance prompt signs to guide the miners to evacuate safely. For example, when the risk of fissure expansion or roof instability is detected, the horizontal roadway is preferentially selected for evacuation to avoid miners being injured by vertical falling rocks; when the water inrush disaster is detected, the uphill evacuation path is preferentially selected to prevent miners from entering low-lying water accumulation areas; when the risk of rock burst or collapse is detected, the shortest path is preferentially selected for evacuation to reduce the residence time of miners in high-stress areas. The intelligent guidance module provides evacuation instructions through LED indicator lights, voice broadcasts, miners' intelligent terminals, etc. The LED indicator lights represent low-risk, medium-risk and high-risk areas in green, yellow and red respectively. The red light flashes and is accompanied by a warning sign to remind the miners to evacuate immediately; the voice broadcast plays the corresponding warning information according to the risk type, such as reminding the rock formation to be unstable when the fissure expands, suggesting to evacuate along the uphill route when there is a water inrush risk, and requiring the miners to evacuate along the shortest path when there is a rock burst risk. In addition, the intelligent terminal device worn by the miners will synchronously display the evacuation direction and provide vibration feedback to prompt the miners to act according to the planned path. Through the above process, the real-time safety guidance of the miners is realized. After receiving the audible and visual alarm and the instructions of the intelligent guidance system, the miners can quickly and orderly evacuate the dangerous area according to the prompt signs, ensuring personal safety. This mechanism not only improves the efficiency of disaster response but also minimizes the risk of mine accidents.

[0047] In a possible implementation manner, based on the multiple historical risk feature distributions, a risk trend prediction is performed on the rock formation risk feature distribution, and a predicted disaster critical condition is output. The method includes:

[0048] Interactively obtain multiple sample critical defect features of the multiple types of rock stratum defects; based on the overlap of defect positions, perform defect evolution trend analysis according to the multiple historical risk feature distributions and the rock stratum risk feature distribution to obtain multiple defect evolution rates of the multiple types of rock stratum defects; aggregate the rock stratum risk feature distribution according to the multiple types of rock stratum defects to obtain multiple sets of real-time defect features; perform extreme value calls on the multiple sets of real-time defect features to obtain multiple real-time defect extremes; calculate the evolution time of the multiple real-time defect extremes according to the multiple sample critical defect features and the multiple defect evolution rates to obtain multiple critical working condition times; serialize the multiple critical working condition times, and perform minimum value calls according to the sorting results to output the predicted disaster time; perform defect evolution fitting on the rock stratum risk feature distribution according to the predicted disaster time and the multiple defect evolution rates to output the predicted disaster critical working condition.

[0049] Optionally, when predicting the risk of mine disasters, it is necessary to interactively analyze the historical data of various types of rock formation defects, combine the real-time monitoring data, calculate the evolution trend of the rock formation defects, and finally predict the possible occurrence time and critical state of the disasters. First, obtain multiple sample critical defect characteristics of various types of rock formation defects through interaction. These characteristics include the physical parameters of different types of defects (fractures, faults, stress concentration, water inrush anomalies, etc.) in the critical state, such as the fracture propagation speed, the deformation amplitude of the rock formation, the stress distribution limit value, the water inrush precursor signal, etc. Analyze the typical manifestations of these defects in the critical state through the historical database, extract their temporal characteristics, and establish the corresponding sample critical defect characteristics to provide a benchmark for subsequent trend analysis. After obtaining multiple sample critical defect characteristics, based on the spatial overlap of the defect positions, combine the distributions of multiple historical risk characteristics and the current rock formation risk characteristics to conduct defect evolution trend analysis, and calculate multiple evolution rates of different defect types. For example, by analyzing historical data, the average fracture propagation speed (cm / d), the stress increase in the stress concentration area (MPa / d), the permeability change rate in the water inrush area (% / d), etc. can be calculated. Then, use methods such as regression analysis, deep learning, or finite element simulation to fit the evolution trends of various defects based on historical data and current risk characteristics, so as to obtain the evolution rates of multiple defects and provide data support for disaster time prediction. After that, based on the distribution of rock formation risk characteristics, conduct feature aggregation on various types of rock formation defects to obtain multiple sets of real-time defect characteristics. For example, integrate all fracture data into one group and fault data into another group. Each group of data includes the position, scale, and current state of the defects. Then, perform extreme value calls on these data to calculate multiple real-time defect extremes. For example, the fracture propagation in a certain area reaches the maximum value, the stress in a certain stress concentration area exceeds the historical maximum value, etc., so as to lock the area and time point most likely to enter the disaster state. Further, based on multiple sample critical defect characteristics and multiple defect evolution rates, calculate the evolution time of multiple real-time defect extremes to predict the time when different defects may evolve to the disaster state. The calculation methods include predictions based on temporal modeling, such as deep learning algorithms like LSTM and GRU. Through these methods, the time required for each defect point to reach the disaster critical state can be calculated, that is, multiple critical working condition times. After obtaining multiple critical working condition times, sequence and sort these time points, and call the minimum value function to determine the earliest possible disaster occurrence time point as the final predicted disaster time. For example, if a certain fracture may reach the critical width in 3 days and a certain water inrush risk area may experience water inrush in 5 days, then 3 days will be selected as the earliest warning time and the corresponding warning mechanism will be activated. Finally, based on the predicted disaster time and multiple defect evolution rates, conduct defect evolution fitting on the current rock formation risk characteristic distribution to output the final predicted disaster critical working condition.This process combines historical data, real-time monitoring data, and disaster evolution models to form a complete risk prediction report, clarifying the possible time, spatial location, impact scope of the disaster, and the corresponding early warning levels, providing accurate disaster prediction support for mine safety management. If the predicted time is short and the risk is high, an emergency warning will be automatically triggered, and the intelligent guidance system will be activated to ensure that miners can evacuate the dangerous area in time.

[0050] Example 2. Based on the same inventive concept as the deep coal mine disaster early warning method based on ground penetrating radar in the foregoing example, as Figure 2 shown, this application provides a deep coal mine disaster early warning system based on ground penetrating radar. The system in the embodiments of this application and the method embodiments are based on the same inventive concept. Among them, the system includes: Equipment layout component 11: Analyze the monitoring coverage of the target monitoring area, and layout the ground penetrating radar equipment according to the analysis results to complete the construction of the ground penetrating radar array; Underground structure scanning component 12: Execute underground structure scanning of the target monitoring area through the ground penetrating radar array to obtain a rock layer data array; Feature recognition component 13: Identify the spatial hierarchical features of the rock layer data array to obtain the rock layer risk feature distribution; Historical data calling component 14: Obtain multiple historical risk feature distributions through historical data calling; Risk trend prediction component 15: Based on the multiple historical risk feature distributions, perform risk trend prediction on the rock layer risk feature distribution and output the predicted disaster critical condition, where the predicted disaster critical condition has a predicted disaster time identifier.

[0051] Furthermore, the equipment layout component 11 further includes:

[0052] Collect the monitoring correlation features of the target monitoring area to determine the monitoring performance indicators, where the monitoring performance indicators include the target monitoring accuracy, equipment resolution, and monitoring coverage; Optimize the layout sites with the target monitoring accuracy and monitoring coverage as constraints to obtain a layout site array; After batch debugging of the ground penetrating radar equipment according to the target monitoring accuracy, refer to the layout site array to layout the ground penetrating radar equipment to complete the construction of the ground penetrating radar array.

[0053] Furthermore, the equipment layout component 11 further includes:

[0054] Digitally process the target monitoring area to obtain a 3D geological model; preset a grid scale, and grid the 3D geological model according to the grid scale to obtain a 3D grid cell model; configure an initial monitoring interval according to the target monitoring accuracy; locate K initial monitoring grid points in the 3D grid cell model according to the initial monitoring interval; take the target monitoring accuracy and the monitoring coverage as constraint conditions, evaluate the monitoring contributions of the K initial monitoring grid points, and screen and update the positions of the grid points according to the evaluation results until the layout site array is output.

[0055] Further, the feature recognition component 13 further includes:

[0056] Pre-construct a rock stratum defect recognition model; load the rock stratum data array into the rock stratum defect recognition model to perform defect recognition and obtain a rock stratum risk defect array; splice the rock stratum risk defect array according to the coverage cross-relationship of the geological radar arrays to obtain the rock stratum risk feature distribution.

[0057] Further, the feature recognition component 13 further includes:

[0058] Interactively obtain multiple sample rock stratum data of multiple rock stratum defect types; perform defect feature marking on the multiple sample rock stratum data to obtain multiple sample defect marking data; use the multiple sample rock stratum data and multiple sample defect marking data as training data to construct multiple rock stratum defect recognition sub-models; interactively obtain multiple defect accident frequencies of the multiple rock stratum defect types; complete the construction of the rock stratum defect recognition model by paralleling the multiple rock stratum defect recognition sub-models and allocating operation computing power according to the multiple defect accident frequencies.

[0059] Further, the feature recognition component 13 further includes:

[0060] Call the first local rock stratum data from the rock stratum data array; after inputting the first local rock stratum data into the rock stratum defect recognition model, perform parallel defect recognition through multiple rock stratum defect recognition sub-models in the rock stratum defect recognition model to obtain multiple local rock stratum defect spaces; perform feature fusion on the multiple local rock stratum defect spaces to obtain a first rock stratum risk defect; by analogy, load the rock stratum data array into the rock stratum defect recognition model, perform defect recognition, and store the recognition results based on the array structure to obtain the rock stratum risk defect array.

[0061] Further, the risk trend prediction component 15 further includes:

[0062] Decompose the predicted disaster critical working conditions to obtain multiple disaster warning index thresholds; identify the spatial hierarchical features of the monitoring data array transmitted in real time by the ground penetrating radar array to obtain the real-time risk feature distribution; traverse the real-time risk feature distribution with the multiple disaster warning index thresholds; when any of the real-time risk feature distributions meets the multiple disaster warning index thresholds, output the associated acoustic and optical warnings according to the index response type.

[0063] Further, the risk trend prediction component 15 further includes:

[0064] Send the associated acoustic and optical warnings to the intelligent guidance module to semi-automatically activate the intelligent guidance module; the intelligent guidance module starts the associated guidance indication according to the associated acoustic and optical warnings.

[0065] Further, the risk trend prediction component 15 further includes:

[0066] Interactively obtain multiple sample critical defect features of the multiple rock layer defect types; based on the overlap of defect positions, perform defect evolution trend analysis according to the multiple historical risk feature distributions and the rock layer risk feature distributions to obtain the multiple defect evolution rates of the multiple rock layer defect types; aggregate the rock layer risk feature distributions according to the multiple rock layer defect types to obtain multiple groups of real-time defect features; perform extreme value calls on the multiple groups of real-time defect features to obtain multiple real-time defect extremes; calculate the evolution time of the multiple real-time defect extremes according to the multiple sample critical defect features and the multiple defect evolution rates to obtain multiple critical working condition times; serialize the multiple critical working condition times, and perform minimum value calls according to the sorting results to output the predicted disaster time; perform defect evolution fitting on the rock layer risk feature distribution according to the predicted disaster time and the multiple defect evolution rates to output the predicted disaster critical working conditions.

[0067] It should be noted that the above sequence of embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above describes specific embodiments of this specification. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-tasking and parallel processing are also possible or may be advantageous.

[0068] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0069] This specification and the accompanying drawings are merely illustrative of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. A deep coal mine disaster early warning method based on geological radar, characterized in that: The method comprises: Conduct monitoring coverage analysis on the target monitoring area, deploy geological radar equipment based on the analysis results, and complete the construction of the geological radar array; Scanning the underground structure of the mine in the target monitoring area by using the geological radar array to obtain a rock formation data array; By performing spatial hierarchical feature recognition on the rock formation data array, a rock formation risk feature distribution is obtained; By calling historical data, multiple historical risk feature distributions are obtained; Based on the multiple historical risk characteristic distributions, risk trend prediction is performed on the rock formation risk characteristic distribution, and a predicted disaster critical condition is output, wherein the predicted disaster critical condition has a predicted disaster time mark.

2. The deep coal mine disaster early warning method based on geological radar according to claim 1, characterized in that: According to the plurality of historical risk characteristic distributions, risk trend prediction is performed on the rock formation risk characteristic distribution, and a predicted disaster critical condition is outputted. Thereafter, the method further comprises: Decomposing the predicted disaster critical conditions to obtain multiple disaster warning indicator thresholds; By performing spatial hierarchical feature recognition on the monitoring data array transmitted back in real time by the geological radar array, a real-time risk feature distribution is obtained; Using the multiple disaster warning indicator thresholds to traverse the real-time risk feature distribution; When any of the real-time risk feature distributions meets the multiple disaster warning indicator thresholds, an associated sound and light warning is output according to the indicator response type.

3. The deep coal mine disaster early warning method based on geological radar according to claim 1 is characterized in that: Performing monitoring coverage analysis on the target monitoring area, and deploying geological radar equipment according to the analysis results to complete the construction of the geological radar array, the method includes: Collect monitoring-related features of the target monitoring area and determine monitoring performance indicators, wherein the monitoring performance indicators include target monitoring accuracy, equipment resolution, and monitoring coverage; Optimizing the deployment sites with the target monitoring accuracy and monitoring coverage as constraints to obtain a deployment site array; After batch debugging of geological radar equipment according to the target monitoring accuracy, geological radar equipment is deployed with reference to the deployment site array to complete the construction of the geological radar array.

4. The deep coal mine disaster early warning method based on geological radar according to claim 3 is characterized in that: The method of optimizing the layout sites based on the target monitoring accuracy and the monitoring coverage as constraints to obtain a layout site array includes: Digitally process the target monitoring area to obtain a geological three-dimensional model; Preset a grid scale, and grid the geological three-dimensional model according to the grid scale to obtain a grid unit three-dimensional model; Configuring the initial monitoring interval according to the target monitoring accuracy; Positioning K initial monitoring grid points in the grid unit three-dimensional model according to the initial monitoring spacing; Taking the target monitoring accuracy and monitoring coverage as constraints, the monitoring contribution of the K initial monitoring grid points is evaluated, and the grid point positions are screened and updated according to the evaluation results until the deployment site array is output.

5. The deep coal mine disaster early warning method based on geological radar according to claim 1, characterized in that: By performing spatial hierarchical feature recognition on the rock formation data array, a rock formation risk feature distribution is obtained, and the method includes: Pre-built rock formation defect identification model; By loading the rock formation data array into the rock formation defect identification model, defect identification is performed to obtain a rock formation risk defect array; According to the coverage intersection relationship of the geological radar array, the rock formation risk defect array is spliced ​​to obtain the rock formation risk characteristic distribution.

6. The deep coal mine disaster early warning method based on geological radar according to claim 5, characterized in that: A rock formation defect recognition model is pre-built, and the method comprises: Interactively obtain multiple sample rock formation data of various rock formation defect types; Performing defect feature identification on the plurality of sample rock formation data to obtain a plurality of sample defect identification data; Using the plurality of sample rock formation data and the plurality of sample defect identification data as training data, constructing a plurality of rock formation defect identification sub-models; Interactively obtaining multiple defect accident frequencies of the multiple rock formation defect types; The construction of the rock formation defect identification model is completed by connecting the multiple rock formation defect identification sub-models in parallel and allocating operating computing power according to the multiple defect accident frequencies.

7. The deep coal mine disaster early warning method based on geological radar according to claim 2, characterized in that: When any of the real-time risk feature distributions meets the multiple disaster warning indicator thresholds, outputting associated sound and light warnings according to the indicator response type, the method includes: Sending the associated sound and light warning to an intelligent guidance module to perform semi-automatic activation of the intelligent guidance module; The intelligent guidance module starts the associated guidance instruction according to the associated sound and light warning.

8. The deep coal mine disaster early warning method based on geological radar according to claim 6, characterized in that: By loading the formation data array into the formation defect identification model to perform defect identification, a formation risk defect array is obtained, the method comprising: Retrieving first local rock formation data from the rock formation data array; After the first local rock formation data is input into the rock formation defect recognition model, multiple rock formation defect recognition sub-models in the rock formation defect recognition model are used to perform parallel defect recognition to obtain multiple local rock formation defect spaces; Performing feature fusion on the multiple local rock formation defect spaces to obtain a first rock formation risk defect; By analogy, the formation data array is loaded into the formation defect identification model, and after defect identification, the identification results are stored based on the array structure to obtain the formation risk defect array.

9. The deep coal mine disaster early warning method based on geological radar according to claim 6, characterized in that: According to the plurality of historical risk characteristic distributions, risk trend prediction is performed on the rock formation risk characteristic distribution, and a predicted disaster critical condition is outputted. The method comprises: interactively obtaining critical defect characteristics of a plurality of samples of the plurality of formation defect types; Based on the overlap of defect positions, defect evolution trend analysis is performed according to the multiple historical risk characteristic distributions and the formation risk characteristic distributions to obtain multiple defect evolution rates of the multiple formation defect types; Aggregating the formation risk feature distribution according to the multiple formation defect types to obtain multiple groups of real-time defect features; Performing extreme value calls on the multiple groups of real-time defect features to obtain multiple real-time defect extreme values; According to the multiple sample critical defect characteristics and multiple defect evolution rates, the multiple real-time defect extreme values ​​are calculated to obtain multiple critical operating conditions; Serialize the multiple critical working condition times, perform minimum value call according to the sorting result, and output the predicted disaster time; According to the predicted disaster time and multiple defect evolution rates, defect evolution fitting is performed on the rock formation risk characteristic distribution, and the predicted disaster critical condition is output.

10. The deep coal mine disaster early warning system based on geological radar is characterized by: The system is used to execute the deep coal mine disaster early warning method based on geological radar according to any one of claims 1 to 9, comprising: Equipment deployment component: Conduct monitoring coverage analysis on the target monitoring area, and deploy geological radar equipment based on the analysis results to complete the construction of the geological radar array; Underground structure scanning component: performing underground structure scanning of the mine in the target monitoring area through the geological radar array to obtain a rock formation data array; Feature recognition component: obtaining the risk feature distribution of the rock formation by performing spatial hierarchical feature recognition on the rock formation data array; Historical data calling component: obtains multiple historical risk feature distributions through historical data calling; Risk trend prediction component: based on the multiple historical risk characteristic distributions, perform risk trend prediction on the rock formation risk characteristic distribution, and output predicted disaster critical conditions, wherein the predicted disaster critical conditions have a predicted disaster time mark.

Images