Deep well rock burst early warning system and method based on multi-dimensional monitoring

Abstract

The invention discloses a deep well rock burst early warning system and method based on multi-dimensional monitoring, and belongs to the technical field of deep well rock burst early warning. According to the method, multi-dimensional data such as stress, strain and microseism are collected through a monitoring network, and an aligned multi-source data set is obtained through space-time registration; after dynamic noise suppression processing matched with physical characteristics is adopted, strong correlation characteristics are screened through mutual information entropy; frequency domain, time domain and time-frequency domain features are extracted through principal component extraction and phase-space reconstruction, and a multi-dimensional state space data set is formed; and inputting the prediction model to obtain a danger level and trigger a corresponding early warning signal, and finally dynamically adjusting the monitoring network layout and prevention and control measures based on the early warning signal. According to the method, the early warning accuracy and real-time performance are improved, and effective technical support is provided for deep well rock burst prevention and control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep well rock burst early warning, and in particular to a deep well rock burst early warning system and method based on multi-dimensional monitoring. BACKGROUND

[0002] With the increasing depletion of shallow mineral resources, mineral exploitation gradually extends to deep wells, and the geomechanical environment faced by deep well tunnels and stope is increasingly complex. Rock burst, as a sudden and destructive dynamic disaster, seriously threatens the safety of underground workers and production equipment, so building an efficient rock burst early warning system has become a core requirement for safe mining in deep wells.

[0003] Existing rock burst early warning technologies are mostly based on a single or a few monitoring parameters, and determine the dangerous state by setting fixed thresholds or empirical formulas. Some technologies attempt to integrate multi-source data, but mainly rely on simple data splicing or weighted analysis, lacking a systematic spatiotemporal correlation processing mechanism. At the same time, existing early warning models mostly use traditional statistical methods or single machine learning algorithms, which have limited adaptability to complex and dynamic environments in deep wells.

[0004] The current technology has three problems: first, the monitoring dimension is single, which is difficult to fully reflect the multi-physical field evolution characteristics of rock burst incubation; second, the data processing method lacks pertinence and does not consider the physical property differences of different types of monitoring data, resulting in poor noise suppression effect and insufficient feature extraction; third, the linkage of early warning and prevention and control measures is insufficient, which cannot dynamically adjust the monitoring strategy according to the real-time early warning results, and it is difficult to realize the closed-loop management from early warning to prevention and control. SUMMARY

[0005] Therefore, the present application provides a deep well rock burst early warning system and method based on multi-dimensional monitoring to solve at least one of the above technical problems.

[0006] To achieve the above-mentioned purpose, in a first aspect, a deep well rock burst early warning method based on multi-dimensional monitoring is provided, which includes the following steps: Using the monitoring network arranged in the deep well tunnel and surrounding rock, real-time collection of multi-dimensional monitoring data including stress value, strain value, microseismic event parameter, acoustic signal, temperature data, radon concentration value, electromagnetic radiation intensity and tunnel surface three-dimensional deformation data is performed; The multi-dimensional monitoring data is spatiotemporally registered to obtain a spatiotemporally aligned multi-source monitoring data set; The multi-source monitoring data set is subjected to dynamic noise suppression based on physical properties to obtain a noise-suppressed data set; Features related to rock burst intensity are selected from the noise-suppressed data set by mutual information entropy algorithm to obtain a feature-optimized data set; extracting frequency domain features, time domain features and time-frequency domain features from the feature-optimized data set by a principal component extraction method and a phase space reconstruction method, to form a multi-dimensional state space data set; inputting the multi-dimensional state space data set into a pre-trained rock burst danger level prediction model to obtain a rock burst danger level; triggering a warning signal of a corresponding level according to the rock burst danger level; dynamically adjusting the layout of the monitoring network and the control measures based on the warning signal of the corresponding level.

[0007] In a second aspect, a deep well rock burst early warning system based on multi-dimensional monitoring is provided, which includes: A data acquisition module, which includes a monitoring network arranged in a deep well tunnel and surrounding rock, for acquiring multi-dimensional monitoring data including stress value, strain value, microseismic event parameter, acoustic signal, temperature data, radon concentration value, electromagnetic radiation intensity and tunnel surface three-dimensional deformation data in real time; A time-space registration module, for time-space registration of the multi-dimensional monitoring data to obtain a multi-source monitoring data set aligned in time and space; A noise suppression module, for dynamic noise suppression of the multi-source monitoring data set based on physical characteristics to obtain a noise-suppressed data set; A feature screening module, for screening features related to rock burst intensity from the noise-suppressed data set by a mutual information entropy algorithm to obtain a feature-optimized data set; A feature extraction module, for extracting frequency domain features, time domain features and time-frequency domain features from the feature-optimized data set by a principal component extraction method and a phase space reconstruction method, to form a multi-dimensional state space data set; A level prediction module, which includes a pre-trained rock burst danger level prediction model, for inputting the multi-dimensional state space data set into the model to obtain a rock burst danger level; A warning triggering module, for triggering a warning signal of a corresponding level according to the rock burst danger level; A dynamic adjustment module, for dynamically adjusting the layout of the monitoring network and the control measures based on the warning signal of the corresponding level.

[0008] The above technical solution has the following beneficial technical effects: The technical scheme forms a complete closed loop through multi-dimensional monitoring, space-time registration, noise suppression, feature screening and extraction, model prediction and dynamic adjustment, realizes high precision, real-time and intelligent of rock burst early warning, multi-dimensional data acquisition covers stress, strain, microseismic and other physical field information, combined with space-time registration technology to ensure data space-time consistency and improve the adaptability of the early warning system to complex geological environment, dynamic noise suppression based on physical characteristics and mutual information entropy feature screening effectively remove interference and focus on key features, improve the early warning accuracy, principal component extraction combined with phase space reconstruction, mine the frequency domain, time domain and nonlinear characteristics of data, enhance the identification ability of the system to rock burst precursor, output graded early warning signal through the prediction model, and adjust the monitoring layout and control measures in linkage, form a dynamic response mechanism, greatly shorten the early warning response time, reduce the false alarm rate, and provide reliable safety guarantee for deep well mining. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 is the overall flowchart of the deep well rock burst early warning method based on multi-dimensional monitoring of the embodiment of the present application; Figure 2 is the specific flowchart of step S10 of the embodiment of the present application; Figure 3 is the specific flowchart of step S20 of the embodiment of the present application; Figure 4 is the specific flowchart of step S30 of the embodiment of the present application; Figure 5 is the specific flowchart of step S40 of the embodiment of the present application; Figure 6 is the specific flowchart of step S50 of the embodiment of the present application; Figure 7 is the specific flowchart of step S70 of the embodiment of the present application; Figure 8 is the specific flowchart of step S80 of the embodiment of the present application; Figure 9 is the functional block diagram of the deep well rock burst early warning system based on multi-dimensional monitoring of the embodiment of the present application; Figure 10 is the structural schematic diagram of the computer system of the embodiment of the present application. DETAILED DESCRIPTION

[0010] The following description of exemplary embodiments of the present invention is made in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0011] like Figure 1 As shown, this embodiment provides a deep well rock burst early warning method based on multi-dimensional monitoring, which includes the following steps: S10: Utilizing the monitoring network deployed in deep well tunnels and surrounding rocks, multi-dimensional monitoring data including stress values, strain values, microseismic event parameters, acoustic wave signals, temperature data, radon gas concentration values, electromagnetic radiation intensity, and three-dimensional deformation data of the tunnel surface are collected in real time.

[0012] Specifically, a monitoring network deployed in deep mine tunnels and surrounding rock allows for real-time collection of multi-dimensional monitoring data. In a deep mining tunnel at a certain mine, monitoring sections are set up every 50 meters along the tunnel's axis. Stress and strain sensors are deployed on the tunnel's roof, sides, and floor at each section. A microseismic monitoring system and acoustic wave monitoring device are deployed every 100 meters. Temperature sensors and radon gas concentration sensors are arranged in a grid pattern within a 500-meter radius of the tunnel. An array of laser displacement sensors is deployed along the entire length of the tunnel to monitor three-dimensional surface deformation. Electromagnetic radiation monitoring devices are also installed in key areas. All sensors transmit collected stress and strain values, microseismic event parameters, acoustic wave signals, temperature data, radon gas concentration, electromagnetic radiation intensity, and three-dimensional tunnel surface deformation data in real time to a ground monitoring center via a combination of ZigBee wireless transmission and Industrial Ethernet.

[0013] S20: Performing spatiotemporal registration on the multi-dimensional monitoring data to obtain a spatiotemporally aligned multi-source monitoring data set.

[0014] Specifically, this step performs spatio-temporal registration of multi-dimensional monitoring data. A three-dimensional geological model spatio-temporal reference coordinate system is established with the roadway entrance as the origin (0, 0, 0), the roadway axial direction as the X-axis, the vertical roadway direction as the Y-axis, and the vertical roadway floor as the Z-axis. Time stamp calibration is performed on each sensor data, and a GPS time service system is used to ensure that the time synchronization error of all sensors is less than 10 ms. For asynchronous data, a cubic spline interpolation method is used for time alignment. A coordinate conversion matrix is constructed to convert the original coordinates of each sensor to the reference coordinate system, for example, a spherical coordinate conversion is used for microseismic sensors, and an affine transformation is used for displacement sensors. A Kriging interpolation method is used for grid processing of spatial discrete point data to form a 5m x 5m x 5m three-dimensional data field, and the spatio-temporal registration error is calculated. When the position error is less than 0.5m and the time error is less than 5ms, the spatio-temporally aligned multi-source monitoring data set is output.

[0015] S30: performing dynamic noise suppression based on physical characteristics on the multi-source monitoring data set to obtain a noise-suppressed data set.

[0016] Specifically, this step performs dynamic noise suppression based on physical characteristics on the multi-source monitoring data set. For mechanical data, a stress-strain trend prediction model is established, and when the deviation between the measured value and the predicted value exceeds 3%, a drift compensation algorithm is started. For microseismic and acoustic signals, wavelet packet decomposition combined with adaptive threshold denoising is used, and the decomposition layer number is dynamically adjusted according to the signal energy distribution. For temperature and radon concentration data, the baseline value is calculated through a sliding window, and when the fluctuation exceeds the baseline ±0.5℃ or ±20Bq / m³, correction is performed. For electromagnetic radiation and displacement data, independent component analysis is used to separate interference signals. The signal-to-noise ratio and feature retention degree before and after denoising are calculated in real time, and when the signal-to-noise ratio is ≥15dB and the feature retention degree is ≥90%, the noise-suppressed data set is output.

[0017] S40: selecting features related to rock burst intensity from the noise-suppressed data set through a mutual information entropy algorithm to obtain a feature-optimized data set.

[0018] Specifically, this step selects features related to rock burst intensity through a mutual information entropy algorithm. A mutual information entropy calculation model of features and rock burst danger is constructed, for example: ; where X is each monitoring feature, and Y is the rock burst danger level. The mutual information entropy values of the stress change rate, microseismic energy release rate, and acoustic wave velocity attenuation rate are calculated to be greater than 0.7 (feature contribution threshold), and these strongly correlated features are retained. The mutual information entropy redundancy between features is further calculated, and when the similarity of the mutual information entropy values of two features exceeds 0.85 (feature redundancy threshold), the feature with more explicit physical meaning is retained, and finally an optimized data set containing 12 key features is formed.

[0019] S50: Extracting frequency domain features, time domain features and time-frequency domain features from the feature-optimized dataset by principal component extraction method and phase space reconstruction method to form a multi-dimensional state space dataset.

[0020] Specifically, multi-dimensional features are extracted by principal component extraction method and phase space reconstruction method. A covariance matrix is constructed for 12 features, eigenvalue decomposition is performed, and the first 5 principal components with cumulative variance contribution rate exceeding 85% (principal component extraction threshold) are extracted. The embedding dimension m=6 is calculated by Cao method, the time delay τ=5 is determined by mutual information method, and the phase space is reconstructed. Fourier transform is performed on each principal component to extract frequency domain features (such as dominant frequency, frequency band energy distribution), time domain statistical features (mean, variance, skewness, etc.) are calculated, and db4 wavelet is used for 5-layer decomposition to extract time-frequency domain features, finally forming a multi-dimensional state space dataset containing 35 features.

[0021] S60: Inputting the multi-dimensional state space dataset into a pre-trained rock burst danger level prediction model to obtain a rock burst danger level.

[0022] Specifically, the multi-dimensional state space dataset is input into the pre-trained rock burst danger level prediction model. The XGBoost algorithm (Extreme Gradient Boosting) is used to build the prediction model, and the training set contains the historical monitoring data of the mine in the past 3 years and the corresponding rock burst event records. The model is trained in 5-fold cross-validation mode, the learning rate is adjusted to 0.1, the maximum depth of the tree is 6, and the minimum sample weight is 3. The model output is divided into four danger levels: level I (safe), level II (low danger), level III (medium danger), and level IV (high danger). The model performance is evaluated by ROC (Receiver Operating Characteristic) curve, and the AUC (Area Under the Curve) value reaches 0.92. The real-time multi-dimensional state space data is input into the model to obtain the current rock burst danger level.

[0023] S70: Triggering a warning signal of the corresponding level according to the rock burst danger level.

[0024] Specifically, according to the rock burst danger level, a warning signal of the corresponding level is triggered. When the model output is level I, the system displays a green safe state; when the output is level II, a yellow warning is triggered, prompting to strengthen monitoring; when the output is level III, an orange warning is triggered, automatically generating a prevention and control measure suggestion report; when the output is level IV, a red warning is triggered, immediately cutting off the power supply in the dangerous area, starting the sound and light alarm system, and notifying relevant personnel to evacuate urgently through short message, APP push and other ways.

[0025] S80: Based on the early warning signal of the corresponding level, dynamically adjust the layout and prevention and control measures of the monitoring network.

[0026] Specifically, the monitoring network layout and prevention and control measures are dynamically adjusted based on the early warning signal. When the early warning signal is level I (safe), the normal layout and data collection frequency of the monitoring network (for example, 5 minutes / time) are maintained, and only daily equipment inspection and maintenance are performed; when it is level II (low risk), the basic layout of the core monitoring area and the secondary monitoring area is maintained, and the data collection frequency of the peripheral monitoring area is increased by 20%, and all sensor working conditions are checked simultaneously to ensure that there are no faults; when it is level III (medium risk), the sampling frequency of the secondary monitoring area sensors is increased by 50% (for example, from 10 minutes / time to 4 minutes / time), and the backup stress sensors and acoustic wave monitoring devices in the core monitoring area are activated to enhance data redundancy, and at the same time, support strength reinforcement is implemented (for example, the anchor preload is increased to the design level). At least one of the prevention and control measures of: strengthening the support strength, directional stress unloading (drilling in high-stress areas across the entire region) or oriented stress unloading (drilling in three high-stress points for unloading); when it is Level IV (high risk), increase the spatial distribution density of sensors in the core monitoring area to twice the original density (for example, from 1 per 10 meters to 1 per 5 meters), increase the frequency of data acquisition in the entire network to real-time acquisition (1 second / time), and simultaneously initiate a combination of prevention and control measures of support strength enhancement, oriented stress unloading (drilling in high-stress areas across the entire region), and grouting reinforcement of surrounding rocks (using ultra-fine cement slurry). All adjustment parameters and the implementation effects of the prevention and control measures (such as stress reduction and surrounding rock deformation rate) are fed back to the monitoring network in real time to form a dynamic optimization closed loop.

[0027] The advantage of this technical solution is that after the multi-dimensional monitoring data is aligned in time and space, the intrinsic correlation of different physical field information is deeply explored, realizing multi-dimensional cross-validation of the early precursors of rock burst, and can better capture the complex signal characteristics of the disaster incubation stage compared with traditional single-dimensional monitoring; dynamic noise suppression based on physical characteristics can filter out interference in a targeted manner while retaining the weak features related to rock burst in various types of data, which are often misjudged as noise by traditional noise reduction methods, thereby effectively reducing false alarms and missed alarms; the dynamic closed-loop linkage between early warning signals and monitoring networks and prevention and control measures enables the system to autonomously optimize parameters through continuous feedback, and maintain stable early warning reliability in complex and changeable deep well geological environments; this full-process collaboration not only improves the accuracy and timeliness of early warnings, but also forms an active response mechanism in disaster prevention and control, and enhances the comprehensive prevention and control capabilities of rock burst disasters.

[0028] like Figure 2 As shown, step S10 may specifically include the following steps: S101: Zoning planning, which is based on the geomechanical parameters of deep well tunnels and the historical risk zoning of rock burst to determine the core monitoring area, secondary monitoring area and peripheral monitoring area of ​​the monitoring network.

[0029] In implementation, the geological mechanics parameters of the roadway, such as the ground stress distribution, surrounding rock strength, coal seam thickness, fault distribution, etc., are collected, combined with the historical data of the position, intensity and influence range of the rock burst event in the past ten years, and the risk matrix method is used for risk zoning. Among them, the core monitoring area is zoned as the area where the historical rock burst frequently occurs and the ground stress value exceeds 80% of the limit strength of surrounding rock (such as the fault intersection zone and the coal seam thickness mutation section); the secondary monitoring area is the area within 100-300 meters outside the core area, with a ground stress value of 50%-80% of the limit strength of surrounding rock; the peripheral monitoring area is the area outside the secondary area, with a ground stress value less than 50% of the limit strength of surrounding rock, and the roadway extension and surrounding area, forming a gradient monitoring coverage from high risk to low risk.

[0030] S102: Sensor deployment, which sets stress sensors, strain sensors, microseismic monitoring systems and acoustic wave monitoring devices in the core monitoring area, sets temperature sensors, radon concentration sensors and electromagnetic radiation monitoring devices in the secondary monitoring area, and sets laser displacement sensor arrays in the peripheral monitoring area.

[0031] In implementation, the core monitoring area adopts a dense arrangement mode, and one group of stress sensors and strain sensors are set every 5 meters in the roadway roof, two sides and floor, and one set of microseismic monitoring system (a triangle array composed of three geophones) and acoustic wave monitoring device (the emission and receiving ends are arranged at the corresponding positions of the two sides of the roadway) are arranged every 20 meters; the secondary monitoring area arranges temperature sensors and radon concentration sensors on the sidewall of the roadway every 30 meters, sets one electromagnetic radiation monitoring device every 50 meters, and the monitoring points avoid water dripping and support structure shielding area; the peripheral monitoring area sets one group of laser displacement sensor arrays every 100 meters along the strike of the roadway, each group contains three laser emitters (aiming at the roof, left side and right side of the roadway respectively) and one receiving terminal, ensuring the coverage of three-dimensional deformation monitoring range of the roadway surface.

[0032] In some embodiments, the stress sensors of the core monitoring area are arranged in a quincunx pattern along the two sides and roof of the roadway and installed at a depth suitable for monitoring the stress of the rock mass, for real-time monitoring of the static stress changes of the rock mass, and connected to the redundant port of the industrial Ethernet switch through a shielded cable; the strain sensors are arranged on the surface of the roadway support structure and the shallow part of the surrounding rock, and alternately arranged with the stress sensors at intervals according to the monitoring accuracy requirements, for monitoring the strain response of the surrounding rock and the support structure, and realizing parallel data transmission through the same industrial Ethernet channel; the detectors of the microseismic monitoring system are arranged in groups along the axial direction of the roadway according to the monitoring resolution requirements, and multiple detectors in each group are distributed in a spatial distribution manner on the floor and the two sides of the roadway, installed at a depth suitable for capturing microseismic signals, for capturing microseismic events generated by rock mass rupture, and connected to the microseismic data acquisition host with time synchronization function through an optical fiber link; the transmitting end and the receiving end of the acoustic wave monitoring device are installed on the opposite two sides of the roadway, respectively, with a spacing set according to the acoustic wave propagation characteristics test requirements, the transmitting end is embedded in the rock mass at a certain depth, and the receiving end is arranged on the support surface, for acquiring the acoustic wave propagation velocity and attenuation characteristics of the rock mass, and sharing the industrial Ethernet redundant channel with the microseismic monitoring system; the temperature sensors of the secondary monitoring area are installed at a suitable height on the sidewall of the roadway and arranged at intervals according to the environmental temperature field monitoring requirements, connected to the mine Ethernet node in series through a mine flame-retardant cable, for monitoring the changes of the environmental temperature field; the radon concentration sensor is hung at a proper position below the roof of the roadway and arranged alternately with the temperature sensor, and the gas concentration data is transmitted through the same mine Ethernet channel; the electromagnetic radiation monitoring device is fixed at a suitable height on the non-mining side of the roadway, the monitoring probe faces the center of the roadway, and is connected to the regional data concentrator through a shielded twisted pair; the laser displacement sensor array of the peripheral monitoring area is arranged along the strike of the roadway according to the deformation monitoring range requirements, each monitoring station contains multiple laser emitters, respectively pointing to the key deformation monitoring points of the roof and the two sides of the roadway, the sensor body is fixed to the stable section of the roadway through an explosion-proof support, the laser reflection target is set at the corresponding monitoring point, and the sensor is connected to the ad hoc network gateway through a wireless radio frequency module, for real-time capturing of the three-dimensional deformation of the roadway surface.

[0033] S103: setting parameters, which set the collection parameters of multi-dimensional monitoring data, including sampling frequency, trigger condition and range.

[0034] In specific implementation, the sampling frequency of the stress sensor and the strain sensor in the core monitoring area is set to 10 Hz, the trigger condition is that the stress change rate exceeds 5% / min, and the high-frequency collection is automatically started (up to 100 Hz) when the trigger condition is met, and the range covers 0-100 MPa; the sampling frequency of the microseismic monitoring system is set to 1 kHz, and the trigger threshold is that the vibration amplitude exceeds 0.01 mm / s; the sampling frequency of the acoustic wave monitoring device is set to 500 Hz, and the range corresponds to the acoustic wave velocity of 300-6000 m / s. The sampling frequency of the temperature sensor in the secondary monitoring area is set to 1 Hz, and the range covers-5-60℃; the sampling frequency of the radon concentration sensor is set to 0.1 Hz (once every 10 seconds), and the range is 0-1000 Bq / m³; the sampling frequency of the electromagnetic radiation monitoring device is set to 5 Hz, and the range is 0-1000 μV / m. The sampling frequency of the laser displacement sensor array in the peripheral monitoring area is set to 0.5 Hz, the trigger condition is that the single deformation amount exceeds 0.1 mm, and the range covers 0-500 mm.

[0035] S104: Data transmission, which constructs a hierarchical data transmission link based on the importance level of the monitoring area, wherein the data of the core monitoring area is transmitted through an industrial Ethernet channel with redundant backup, the data of the secondary monitoring area is transmitted through a mine Ethernet channel, and the data of the peripheral monitoring area is transmitted through a wireless ad hoc network aggregation, realizing real-time transmission of the multi-dimensional monitoring data.

[0036] In specific implementation, the core monitoring area adopts a dual-line redundant industrial Ethernet channel, the main line is an underground gigabit optical network, and the standby line is a fiber ring network, and a priority scheduling mechanism is used to ensure that the stress, microseismic and other data are transmitted preferentially, and the transmission delay is controlled within 100 ms; the data of the secondary monitoring area is transmitted through a mine Ethernet switch to form a star network, a data frame queuing mechanism is used to avoid transmission conflicts, and the transmission delay is controlled within 500 ms; the laser displacement sensor array in the peripheral monitoring area is connected through a Zigbee wireless ad hoc network, each sensor acts as a network node, data is aggregated to a regional gateway through multi-hop routing and then connected to the backbone network, the transmission period is dynamically adjusted according to the sampling frequency, and the data is ensured to be uploaded completely.

[0037] S105: Data labeling, which synchronizes the time stamp and marks the geological coordinates of the multi-dimensional monitoring data.

[0038] In specific implementation, the Beidou timing system is used to uniformly time all sensors, the timestamp accuracy is controlled within 1 ms, and the time reference of data collected by different devices is ensured to be consistent; a coordinate system based on a three-dimensional geological model of a roadway is used to assign a unique geological coordinate to each sensor (for example, a stress sensor in a core area is marked as X=1250 m, Y=25 m, and Z=5 m, corresponding to a position of 1250 m in the axial direction of the roadway, 25 m on the left side, and 5 m above the floor), and the coordinate information is embedded in the metadata of the corresponding monitoring data to achieve accurate association of the data and the spatial position.

[0039] The technical effects of the technical solution are that, based on the monitoring area division of the geological mechanics parameters and the historical risk zoning, dynamic focusing of the monitoring resources is achieved, high-density key parameter monitoring in the core area is combined with supplementary parameter monitoring in the secondary area and the peripheral area, resource redundancy is avoided, and the multi-physical field characteristics of the rockburst incubation are completely covered; the construction of the hierarchical data transmission link not only ensures the real-time performance of the key data through redundant transmission in the core area, but also reduces data congestion through adaptive transmission in the secondary area and the peripheral area, so that the stability of the whole-system data transmission is improved compared with a single transmission mode; the integration of the time stamp synchronization and the geological coordinate marking on the hierarchical monitoring data strengthens the spatiotemporal correlation of the data in different areas and of different types, so that cross-area precursor signal coupling analysis in subsequent data processing is possible.

[0040] As shown in Figure 3 , step S20 can specifically include the following steps: S201: Establish a three-dimensional geological coordinate system (origin and three-axis definition), which is established based on a three-dimensional geological model of a deep well roadway. The spatiotemporal reference coordinate system takes the starting point of the roadway as the origin, the axial direction of the roadway as the X axis, the direction perpendicular to the roadway as the Y axis, and the direction perpendicular to the roadway floor as the Z axis, and is associated with the absolute geological coordinate.

[0041] In some embodiments, a local coordinate system is formed with the starting point of the roadway excavation (corresponding to the wellhead coordinates X=523000 m, Y=3762000 m, and Z=-850 m) as the origin (0, 0, 0), the roadway excavation direction (30° north by east) as the X axis, the direction perpendicular to the roadway (30° northwest by west) as the Y axis, and the direction perpendicular to the roadway floor upward as the Z axis. Through three high-level control points (coordinate accuracy better than 0.1 m) known in the mining area, a seven-parameter coordinate conversion method (including three translation parameters, three rotation parameters, and one scale parameter) is used to associate the local coordinate system with the national 2000 geodetic coordinate system, to realize real-time conversion of the local coordinate and the absolute geological coordinate.

[0042] S202: Time calibration (unified time axis and asynchronous interpolation), which timestamps the multi-dimensional monitoring data, unifies the acquisition time of various monitoring devices for collecting the multi-dimensional monitoring data to the standard time axis of the space-time reference coordinate system, and performs time dimension alignment on time asynchronous data using linear interpolation method.

[0043] In some embodiments, all monitoring devices access the mine unified time system (based on Beidou satellite time service, synchronization accuracy ±5ms), and the timestamp is automatically calibrated once per hour. For time asynchronous data caused by communication delay (such as peripheral monitoring data transmitted by ZigBee network), when the time interval of two adjacent valid data points exceeds 2 times the sampling period, the estimated value of the intermediate time is calculated using linear interpolation method: let x1 be the data at time t1, x2 be the data at time t2, and x(t) = x1 + (x2 - x1) × (t - t1) / (t2 - t1) be the interpolation result at time t ∈ (t1, t2), ensuring that all data are aligned in time dimension to the standard time axis of the reference coordinate system (based on Beijing time).

[0044] S203: Space coordinate system conversion (matrix conversion), which constructs a space coordinate conversion matrix of the setting position of each monitoring device based on the space-time reference coordinate system, and converts the original space data of different monitoring devices to three-dimensional coordinates under the space-time reference coordinate system.

[0045] In some embodiments, a conversion model is established for the original coordinate system of different types of monitoring devices: the microseismic monitoring system adopts a spherical coordinate (r, θ, φ) to rectangular coordinate (X, Y, Z) conversion matrix, where X = r × sinθ × cosφ + X0, Y = r × sinθ × sinφ + Y0, Z = r × cosθ + Z0 ((X0, Y0, Z0) is the sensor installation position coordinate); the laser displacement sensor adopts an affine transformation matrix from local plane coordinate to three-dimensional reference coordinate, and the translation, rotation and scaling parameters are calculated through three known control points (such as roadway roof anchor position); the stress sensor directly uses the three-dimensional coordinate of the installation position (obtained by total station measurement) as the original coordinate, without additional conversion. All conversion matrices are optimized by least squares method to ensure that the conversion residual is less than 0.1m.

[0046] S204: Spatial gridding processing (Kriging interpolation method), which performs spatial gridding processing on the monitoring data with uneven spatial distribution using Kriging interpolation method to form a spatial continuous monitoring data field, which is used to integrate the spatial correlation of multi-dimensional monitoring data.

[0047] In some embodiments, the Kriging interpolation method is used for spatial gridding. Based on the three-dimensional coordinates and monitoring values of each monitoring point, a variogram function model is constructed (for example, a spherical model is selected, with a range of 50 m and a base value of 1.2 times the data variance), and interpolation calculation is performed on the core monitoring area with a 2m x 2m x 2m grid, the secondary monitoring area with a 5m x 5m x 5m grid, and the peripheral monitoring area with a 10m x 10m x 10m grid. In the interpolation process, the key areas such as high stress areas and microseismic dense areas are treated with increased sampling point weight to ensure accurate local detail representation, and finally a continuous three-dimensional monitoring data field is formed to intuitively present the spatial distribution of stress, temperature, displacement and other parameters (for example, the spatial overlap relationship between stress concentration areas and temperature anomaly areas).

[0048] S205: Error calculation and verification (time deviation and spatial deviation), which calculates the registration error according to the time synchronization error and the spatial coordinate deviation, wherein the time synchronization error is the deviation value of the time stamp of the calibrated monitoring data and the standard time axis, and the spatial coordinate deviation is the deviation amount of the converted three-dimensional coordinates from the theoretical coordinates of the space-time reference coordinate system; when the registration error exceeds the registration accuracy threshold (i.e. error out of limit), return to step S202 for recalibration, including time calibration or spatial conversion; when the registration error meets the requirements of the registration accuracy threshold (i.e. error meets the standard), output the multi-source monitoring data set aligned in time and space (output the time and space aligned data set).

[0049] In some embodiments, the time synchronization error is calculated using the root mean square error (RMSE, Root Mean Square Error) of the calibrated time stamp and the standard time axis, and the spatial coordinate deviation is calculated using the Euclidean distance mean of the converted coordinates and the theoretical coordinates (generated by the three-dimensional model). The registration accuracy threshold is set as: time synchronization error ≤10ms, spatial coordinate deviation ≤0.5m. When the calculation result exceeds the threshold (for example, a batch of microseismic data has a spatial deviation of 0.8m due to the drift of the geophone), automatically return to step S202 to recalibrate the time or update the conversion matrix; when the error meets the threshold requirements, output the time and space aligned multi-source monitoring data set containing the time stamp (accurate to milliseconds) and the three-dimensional coordinates (accurate to 0.1m), which provides a unified space-time reference for subsequent data processing.

[0050] The beneficial technical effects of the above technical solutions are that the time-space reference coordinate system based on the three-dimensional geological model of the roadway provides a unified reference framework for monitoring data of different types and different positions, avoiding the correlation failure caused by the differences in the coordinate systems of traditional multi-source data; the combination of timestamp calibration and spatial coordinate conversion effectively eliminates the time asynchrony and spatial misplacement problems of data collection, enabling the originally dispersed stress, microseismic, environmental, and other multi-dimensional data to form a relevant association in the same time-space dimension; spatial gridding processing further enhances the spatial continuity of the data, enabling the information of discrete monitoring points to be converted into continuous data fields that can be analyzed globally; and the dynamic checking mechanism of the registration error ensures the controllability of the accuracy of the entire registration process, avoiding invalid data from entering the subsequent processing links.

[0051] As shown in Figure 4 S30 can specifically include the following steps: S301: classifying and identifying noise types, which is based on the physical properties of multi-source monitoring data to classify and identify the noise types corresponding to each dimension of data, wherein the mechanical data includes stress values and strain values, the wave signal includes microseismic event parameters and acoustic signals, the environmental parameters include temperature data and radon concentration values, and the electromagnetic and optical data includes electromagnetic radiation intensity and three-dimensional deformation data of the roadway surface.

[0052] Specifically, the mechanical data (stress values and strain values) is mainly affected by sensor temperature drift and cable contact noise, showing slow trend deviation or instantaneous pulse interference; the wave signal (microseismic event parameters and acoustic signals) is easily affected by roadway mechanical vibration and blasting aftershock interference, showing a wideband random noise superposition feature; the environmental parameters (temperature data and radon concentration values) often produce baseline drift noise due to device heat dissipation and air flow disturbance; and the electromagnetic and optical data (electromagnetic radiation intensity and three-dimensional deformation data of the roadway surface) are easily affected by industrial electromagnetic interference and measurement deviation caused by changes in illumination, showing high-frequency electromagnetic noise or periodic light interference signals. By analyzing the frequency spectrum characteristics and fluctuation rules of the data, a mapping relationship between the noise types and the data categories is established, providing a basis for targeted noise reduction.

[0053] S302: matching a noise reduction algorithm, which uses a noise reduction algorithm matched with the physical characteristics of the data according to the identified noise types, including using a drift compensation algorithm based on physical change trends for the mechanical data, using an adaptive filtering algorithm for the wave signal, using a baseline correction algorithm for the environmental parameters, and using an interference subspace separation algorithm for the electromagnetic and optical data.

[0054] Specifically, for mechanical data, a stress-temperature drift model is established by collecting contemporaneous environmental temperature data, and a least squares method is used to fit the drift curve and perform real-time compensation. For example, when the temperature changes by 1℃, the stress value is linearly corrected by 0.2MPa. For fluctuation signals, an adaptive filter based on the LMS (Least Mean Square) algorithm is used to filter out 50Hz power frequency interference and 200-500Hz mechanical vibration noise while retaining the 2-50Hz effective frequency band of the microseismic signal. For environmental parameters, a sliding window (window size set to 30 minutes) is used to calculate the baseline values of temperature and radon concentration, and baseline correction is performed when the instantaneous value deviates from the baseline by more than 3 times the standard deviation. For electromagnetic and optical data, the interference subspace in the electromagnetic radiation signal is separated by principal component analysis (PCA), and the 150-300MHz frequency band interference generated by industrial equipment is removed. For laser displacement data, wavelet threshold denoising is used to separate the illumination interference component.

[0055] Specifically, the interference subspace separation algorithm is a noise reduction method based on signal subspace decomposition theory. The principle is to decompose the mixed signal contained in electromagnetic and optical monitoring data into useful signal subspace and interference signal subspace through mathematical modeling, and then separate the two types of subspaces through spatial projection operation to filter out noise. Specifically, the algorithm first performs feature analysis on the collected electromagnetic radiation intensity, roadway surface three-dimensional deformation and other data, and identifies the useful signals (such as electromagnetic radiation anomalies caused by rock mass stress changes, characteristic signals of roadway structure deformation) and interference signals (such as device electromagnetic interference, environmental light fluctuations, measurement device noise, etc.) related to rock burst; then through mathematical operations such as eigenvalue decomposition or singular value decomposition, the spatial basis vectors of the two types of signals are determined, and the corresponding subspace model is constructed; finally, through orthogonal projection, the components belonging to the interference subspace in the mixed signal are removed, and only the effective components of the useful signal subspace are retained. This algorithm can specifically separate the complex and dynamically changing interference in electromagnetic and optical data, and is especially suitable for scenarios where multiple sources of interference coexist in deep well environments. It can suppress noise while maximizing the retention of characteristic information related to rock burst, providing high-quality data support for subsequent feature extraction and early warning models.

[0056] S303: Monitor the variability and dynamic parameters. The data variability in the noise suppression process is monitored in real time, and the data variability is used as a feedback to dynamically adjust the parameters of the noise reduction algorithm, so that the suppression intensity is adapted to the dynamic changes of the noise.

[0057] Specifically, the data variability is monitored in real time and the noise reduction parameters are dynamically adjusted. The coefficient of variation (ratio of standard deviation to mean) is used to quantify the data variability. When the coefficient of variation of mechanical data exceeds 10%, the time window for drift compensation is automatically reduced (from 1 hour to 30 minutes) to improve response speed; when the coefficient of variation of fluctuation class signal suddenly increases, the convergence factor of adaptive filter is increased (from 0.01 to 0.05) to enhance noise suppression strength; when the variability of environmental parameters decreases, the threshold for baseline correction is relaxed (from 3 times the standard deviation to 4 times the standard deviation) to avoid overcorrection; when the variability of electromagnetic and optical data is abnormal, the number of principal components retained by PCA is increased (from 3 to 5) to reduce the loss of effective signal and ensure that the suppression strength is dynamically adapted to the real-time changes of noise.

[0058] S304: Calculate the signal-to-noise ratio and feature retention, and judge the threshold. The signal-to-noise ratio and feature retention of the data after noise suppression are calculated. When the signal-to-noise ratio reaches the signal-to-noise ratio threshold and the feature retention reaches the feature integrity threshold, the data set after noise suppression is output; otherwise, return to step S302 to adjust the parameters of the noise reduction algorithm.

[0059] Specifically, the signal-to-noise ratio and feature retention of the data after noise suppression are calculated and threshold judgment is performed. The signal-to-noise ratio is calculated by the ratio of signal power to noise power, where the mechanical data uses the compensated smooth section signal as the reference, and the fluctuation class signal uses the characteristic frequency band of the microseismic event as the effective signal; the feature retention is measured by the correlation coefficient of the key features (such as stress peak, microseismic energy, temperature gradient) of the data before and after noise reduction. When the signal-to-noise ratio reaches the preset signal-to-noise ratio threshold (such as mechanical class ≥ 20 dB, fluctuation class ≥ 15 dB) and the feature retention reaches the feature integrity threshold (such as correlation coefficient ≥ 0.9), the data set after noise suppression is output; if not, return to step S302 to adjust the parameters of the corresponding algorithm (such as filter order, compensation coefficient, etc.) until the threshold requirements are met.

[0060] The technical scheme has the advantages that the classification noise recognition based on the data physical properties avoids the neglect of the differences in noise characteristics of different types of data by the traditional unified noise reduction method, makes the noise suppression more targeted, and accurately distinguishes the drift noise of mechanical data, the interference noise of fluctuation signals, the baseline noise of environmental parameters, and the interference signals of electromagnetic optical data, thereby avoiding the loss of effective features or noise residues caused by unified noise reduction; the differentiated noise reduction algorithm matching the data physical properties ensures that the key features related to rock burst (for example, the mutation trend of stress, the main frequency characteristics of microseismic signals, and the abnormal fluctuations of radon concentration) are completely retained while the noise is suppressed; the dynamic parameter adjustment mechanism can adaptively adjust the noise reduction strength according to the dynamic changes of the noise by monitoring the data variability in real time, thereby effectively dealing with the uncertainty of noise sources in deep well environments; and the dual verification of the signal-to-noise ratio and the feature retention degree strictly controls the quality of the data after noise reduction, thereby avoiding invalid data from entering the subsequent processing link.

[0061] As shown in Figure 5 , step S40 can specifically include: S401: constructing a model, constructing a mutual information entropy calculation model between each feature in the data set after noise suppression and rock burst danger, the mutual information entropy calculation model being based on a mutual information entropy formula in information theory, quantifying the information correlation degree between the feature variable and the rock burst danger level.

[0062] Specifically, when constructing the mutual information entropy calculation model between each feature in the data set after noise suppression and rock burst danger, first, collect the multi-dimensional monitoring data after noise suppression as feature variables X, which include stress values, strain values, microseismic event parameters, sound wave signals, temperature data, radon concentration values, electromagnetic radiation intensity, and roadway surface three-dimensional deformation data, etc. At the same time, sort out the actual occurrence of rock burst and the danger level evaluation results in the corresponding time period as the rock burst danger variable Y, wherein Y can be divided into three levels of low danger, medium danger, and high danger. Based on the mutual information entropy formula I(X;Y)=H(X)+H(Y)-H(X,Y) in information theory, wherein H(X) is the information entropy of the feature variable X, H(Y) is the information entropy of the rock burst danger variable Y, and H(X,Y) is the joint information entropy of X and Y. In actual operation, for each feature variable, for example, stress value, the historical monitoring data thereof is discretized, divided into several intervals, the probability of occurrence of each interval is counted, and H(X) is calculated. Similarly, the probability of occurrence of each danger level of rock burst is counted, and H(Y) is calculated. The joint probability of the feature variable being in a certain interval and the rock burst being in a certain danger level is counted, and H(X,Y) is calculated. Through these calculations, the mutual information entropy calculation model quantifying the information correlation degree between the feature variable and the rock burst danger level is constructed.

[0063] S402: Calculate the mutual information entropy value of each feature by the mutual information entropy calculation model, which represents the prediction contribution of the feature to the rock burst danger.

[0064] Specifically, the mutual information entropy value between each feature in the noise-suppressed dataset and the rock burst danger is calculated by the mutual information entropy calculation model constructed above. For example, for the stress value feature, the interval data after discretization and the corresponding rock burst danger level data are substituted into the model, and the mutual information entropy value of the feature is calculated according to the formula. The larger the mutual information entropy value, the higher the correlation between the feature and the rock burst danger, and the greater the prediction contribution to the rock burst danger. Similarly, the mutual information entropy values of strain value, microseismic event parameters, acoustic signal and other features are calculated in turn, so as to obtain the prediction contribution quantification result of each feature to the rock burst danger.

[0065] S403: Screen the contribution degree feature, which screens the features with mutual information entropy values greater than the feature contribution degree threshold value according to the calculated mutual information entropy values, to form a feature subset related to the rock burst intensity.

[0066] Specifically, according to a large amount of historical monitoring data and rock burst occurrence case analysis, a feature contribution degree threshold value is preset, which can distinguish the features that have significant contribution to the prediction of rock burst danger. For example, the feature contribution degree threshold value is determined to be 0.6 through statistical analysis. Then, the mutual information entropy values of each feature are compared with the threshold value, and the features with mutual information entropy values greater than 0.6 are screened out. Assuming that the mutual information entropy values of stress value, microseismic event parameters and acoustic signal are 0.75, 0.82 and 0.68 respectively, all of which are greater than the threshold value 0.6, then these three features are retained to form a feature subset related to the rock burst intensity.

[0067] S404: Eliminate redundant features, which perform redundancy analysis on the feature subset, eliminate redundant features with similar mutual information entropy values and correlation higher than the feature redundancy threshold value, and obtain the feature-optimized dataset.

[0068] Specifically, the feature subset obtained by screening is analyzed for redundancy. The correlation coefficient between any two features in the feature subset is calculated, such as the correlation between stress value and microseismic event parameters, and the correlation between stress value and acoustic signal, etc. At the same time, the mutual information entropy values of these features are compared. When the mutual information entropy values of two features are similar, and the correlation coefficient between them is higher than the pre-set feature redundancy threshold (for example, 0.8), it indicates that there is a high degree of redundancy between the two features, that is, they provide a high degree of overlap in information when predicting rock burst danger. For example, if the mutual information entropy values of microseismic event parameters and acoustic signal are 0.82 and 0.80 respectively, which are relatively close, and their correlation coefficient is 0.85, which is higher than the feature redundancy threshold 0.8, then one of them is removed (any one can be removed according to actual conditions). In this way, after removing the redundant features, the feature-optimized dataset is obtained.

[0069] The above technical solution quantifies the correlation degree of feature variables and rock burst danger level by constructing a mutual information entropy calculation model, can accurately identify features that have substantial significance for rock burst prediction, and avoid irrelevant or low-contribution data interference; by calculating the mutual information entropy value, the prediction contribution of each feature is clear, providing a quantitative basis for feature screening and ensuring that the feature subset focuses on high correlation parameters; with the help of feature contribution threshold screening and redundancy analysis, redundant features are removed while key information is retained, which not only simplifies the dataset size to reduce the complexity of subsequent model calculation and improve operation efficiency, but also reduces the information overlap between features, avoiding the risk of model overfitting, thereby improving the accuracy and stability of rock burst danger level prediction, providing a more reliable data analysis basis for subsequent early warning signal triggering and prevention and control measure adjustment, and ultimately achieving more accurate and efficient monitoring and early warning of deep rock burst.

[0070] As shown in Figure 6 , step S50 can specifically include the following steps: S501: constructing a covariance matrix for the feature-optimized dataset, and extracting principal components with a variance contribution rate exceeding a pre-set threshold by performing eigenvalue decomposition on the covariance matrix to form a dimension-reduced feature subset.

[0071] Specifically, for the feature-optimized dataset (for example, containing 5 feature variables of stress peak value, microseismic energy, acoustic main frequency, radon concentration gradient, and electromagnetic radiation pulse frequency), first, the mean of each feature variable is calculated, and then a covariance matrix is constructed by the formula, which has a dimension of 5x5, and the matrix elements represent the degree of cooperative change of any two feature variables (for example, the covariance value of stress peak value and microseismic energy is 12.8, reflecting a positive correlation between the two). After eigenvalue decomposition of the covariance matrix, 5 eigenvalues to , respectively, corresponding to the eigenvector representing the direction of the principal component. Set the preset threshold to 85% of the cumulative variance contribution rate, and calculate the cumulative variance contribution rate of the first three eigenvalues (35.2+18.7+9.5) / (35.2+18.7+9.5+4.3+2.1)×100%=89.3%, which meets the threshold requirement, so the first three principal components form the feature subset after dimensionality reduction (containing three principal component variables).

[0072] S502: Based on the phase space reconstruction method, the embedding dimension and the time delay parameter are determined according to the autocorrelation function and the mutual information function of the feature subset after dimensionality reduction, and a high-dimensional phase space is constructed.

[0073] Specifically, based on the time series data (sampling frequency of 1 Hz, time length of 24 hours, a total of 86400 data points) composed of the three principal component variables after dimensionality reduction, the time delay τ is calculated using the autocorrelation function: the autocorrelation coefficient of the principal component time series is calculated, and when the lag time is 5s, the autocorrelation coefficient is close to 0 for the first time, and τ is preliminarily determined as 5s; Then verify by mutual information function, calculate the mutual information value of different lag time, find that when τ=5s, the mutual information value is minimum (0.32), confirm the time delay parameter is 5s. The embedding dimension m is determined by the false nearest neighbor method: start from m=1 and increase the dimension step by step, calculate the proportion of false nearest neighbors in high-dimensional space, when m=5, the proportion of false nearest neighbors drops to 3.2% (lower than 5% threshold), so the embedding dimension is determined as 5. Based on τ=5s and m=5, a high-dimensional phase space is constructed, and each sample point is represented as (x(t), x(t+5), x(t+10), x(t+15), x(t+20)), where x(t) is the value of a principal component at time t.

[0074] S503: In the high-dimensional phase space, the frequency domain feature, the time domain feature and the time-frequency domain feature of each feature are extracted respectively, wherein the frequency domain feature is obtained by Fourier transform, the time domain feature is calculated based on the time series statistical characteristics, and the time-frequency domain feature is extracted by wavelet transform.

[0075] Specifically, in the high-dimensional phase space, frequency domain features are extracted from each principal component time series: time domain signals are converted into frequency domain spectrum through fast Fourier transform (FFT), and the main frequency of principal component 1 is calculated to be 2.3 Hz and the main frequency of principal component 2 is calculated to be 1.8 Hz, which are used as frequency domain features. Time domain features are extracted: based on the statistical characteristics of the time series, the mean value (for example, the mean value of principal component 1 is 12.5 MPa), the peak factor (the peak factor of principal component 2 is 3.8), the kurtosis (the kurtosis of principal component 3 is 4.2), and the root mean square value (the root mean square of principal component 1 is 13.2 MPa) of each principal component are calculated to form time domain features. Time-frequency domain features are extracted: db4 wavelet basis is used for 3-layer wavelet decomposition of the time series, and wavelet coefficients at different scales are obtained. The energy proportion at each scale (for example, the energy proportion at the first scale is 25%, the energy proportion at the second scale is 38%, and the energy proportion at the third scale is 37%) and the wavelet entropy (the wavelet entropy of principal component 1 is 0.62) are calculated as time-frequency domain features.

[0076] S504: The extracted frequency domain features, time domain features, and time-frequency domain features are fused to form a multi-dimensional state space data set.

[0077] Specifically, the extracted frequency domain features (3 main frequency parameters), time domain features (4 statistical quantities of each of the 3 principal components, a total of 12), and time-frequency domain features (4 wavelet parameters of each of the 3 principal components, a total of 12) are fused to form a multi-dimensional state space data set containing 3+12+12=27 feature parameters through feature splicing. Each data sample corresponds to a state point in the high-dimensional phase space, and fully represents the multi-domain feature correlation relationship of the monitored object in the time and space dimensions.

[0078] The mutual information entropy calculation model is constructed to quantify the information correlation degree of each feature and the rock burst danger, accurately select strong correlation features with high contribution degree, effectively exclude irrelevant or low correlation features to interfere with subsequent analysis, improve the pertinence and effectiveness of the feature set, and meanwhile, through redundancy analysis, redundant features with similar mutual information entropy and high correlation are removed, so as to reduce the data dimension, avoid information repetition, reduce the complexity of subsequent feature extraction and model calculation, save computing resources and improve processing efficiency. Finally, the feature-optimized data set can not only retain key information but also simplify the data structure, providing a high-quality data basis for subsequent construction of the multi-dimensional state space data set and accurate output of the prediction model, thereby improving the accuracy and timeliness of the rock burst early warning.

[0079] As shown in Figure 7 , the pre-trained rock burst danger level prediction model is constructed by using a long short-term memory neural network combined with a random forest algorithm optimized by a genetic algorithm.

[0080] The rock burst danger level prediction model can include: a LSTM (Long Short-Term Memory) time series feature extraction and classification layer, configured to encode time series features in a multi-dimensional state space dataset in a bidirectional manner using a bidirectional LSTM network, output a time series feature vector, and perform classification processing on the time series feature vector through a fully connected layer to obtain a probability distribution of rock burst hazard levels in a time series dimension; a feature fusion and dimension reduction layer, configured to concatenate the time series feature vector and static features obtained from a frequency domain feature and a time-frequency domain feature extracted from a data set after feature optimization through principal component extraction and phase space reconstruction, perform weighting through an attention mechanism, and perform dimension reduction using principal component analysis, and output a reduced comprehensive feature vector; a genetic algorithm-optimized random forest classifier, configured to take the reduced comprehensive feature vector as input, optimize random forest hyperparameters using cross-validation accuracy as an adaptive function, and output a probability distribution of rock burst hazard levels in a spatial dimension; a model integration and output layer, configured to fuse the probability distribution of rock burst hazard levels in the time series dimension and the probability distribution of rock burst hazard levels in the spatial dimension through a soft voting method, and output a rock burst hazard level.

[0081] In some embodiments, the LSTM time series feature extraction and classification layer adopts a network structure including an input layer, a bidirectional LSTM layer, and a fully connected layer in specific applications. The bidirectional LSTM layer includes a forward LSTM unit and a backward LSTM unit. The forward LSTM unit encodes time series features in a multi-dimensional state space dataset in chronological order, and the backward LSTM unit encodes in reverse chronological order. The complete time series feature vector is formed by concatenating the output vectors of the two directions, effectively capturing the forward and backward dependencies of the data in the time dimension. The fully connected layer includes two linear transformations and an activation function. The first layer uses a ReLU (Rectified Linear Unit) activation function to perform nonlinear mapping on the time series feature vector, and the second layer outputs a probability distribution of rock burst hazard levels in a time series dimension through a Softmax (normalized exponential function) activation function, with the probability values summing to 1 and each probability value corresponding to the likelihood of the corresponding level.

[0082] In some embodiments, the processing procedure of the feature fusion and dimension reduction layer is as follows: first, static features are screened out from the frequency domain features and time-frequency domain features obtained from principal component extraction and phase space reconstruction of the data set after feature optimization. Such features do not change dynamically with time and mainly reflect the inherent properties of the monitored object; the time series feature vector output by the LSTM time series feature extraction and classification layer is spliced with the above-mentioned static features according to the feature dimension to form a fusion feature matrix. Then, the attention mechanism is introduced, and the attention weight obtained by training is used to weight each feature dimension in the fusion feature matrix, so that the features that have greater impact on the early warning result obtain higher weights, and the representation ability of key information is enhanced. Finally, the principal component analysis method is used to reduce the dimension of the weighted fusion feature matrix, and the principal components with cumulative variance contribution rate exceeding 90% are retained, and redundant information is removed, and a comprehensive feature vector with reduced dimension is output, providing high-quality input for the subsequent classifier.

[0083] In some embodiments, the random forest classifier optimized by genetic algorithm includes two parts: initial model building of random forest and hyperparameter optimization of genetic algorithm. The initial model of random forest contains 500 decision trees, and the training samples of each tree are selected from the training set by bootstrap sampling (bootstrap sampling method), and the feature selection uses the random subset method. In the optimization stage of genetic algorithm, the hyperparameters of random forest (including the number of decision trees, the maximum tree depth, and the minimum leaf node sample size) are used as optimization variables, and the classification accuracy of 5-fold cross-validation is used as the fitness function. The algorithm is iteratively optimized by selection operators (roulette wheel selection), crossover operators (single-point crossover), and mutation operators (random disturbance). When the fitness value does not improve significantly for 10 consecutive generations, the iteration is stopped, and the optimal hyperparameter combination is output. The comprehensive feature vector output by the feature fusion and dimension reduction layer is input into the optimized random forest classifier, and finally the probability distribution of three risk levels in the spatial dimension is obtained, which can effectively reflect the difference in risk degree of the monitoring area in space.

[0084] In some embodiments, the model integration and output layer use soft voting method to realize multi-dimensional result fusion. Specifically, the time series dimension probability distribution output by the LSTM time series feature extraction and classification layer is added to the spatial dimension probability distribution output by the random forest classifier optimized by genetic algorithm according to the corresponding levels, to obtain the cumulative probability value of each risk level. For example, the low-risk probability of the time series dimension is 0.3, the medium-risk is 0.5, and the high-risk is 0.2. The spatial dimension low-risk is 0.2, the medium-risk is 0.4, and the high-risk is 0.4. After accumulation, the low-risk is 0.5, the medium-risk is 0.9, and the high-risk is 0.6. The level with the maximum cumulative probability value is selected as the final rock burst risk level. If the cumulative probability is equal, the higher risk level is selected as the output result to ensure the safety and rigor of the early warning.

[0085] As Figure 8As shown, step S70 can specifically include the following steps: S701: presetting the corresponding relationship between the rock burst danger level and the warning level, wherein the low danger level corresponds to the first warning, the medium danger level corresponds to the second warning, and the high danger level corresponds to the third warning.

[0086] Specifically, when S701 presets the corresponding relationship between the rock burst danger level and the warning level, the parameter setting is completed by the system configuration module: the low danger level is defined as the rock burst occurrence probability being less than 10% and the energy release rate being less than 50 J / h, corresponding to the first warning; the medium danger level is defined as the occurrence probability being 10%-30% and the energy release rate being 50-150 J / h, corresponding to the second warning; and the high danger level is defined as the occurrence probability being more than 30% and the energy release rate being more than 150 J / h, corresponding to the third warning. The corresponding relationship can be configured by an industrial control computer of the underground monitoring center, and a configuration interface provides a visual slider and a numerical input box. The operating personnel can adjust the probability threshold and the energy threshold according to the specific geological conditions of the mine, and after the setting is completed, the parameters are stored in the intrinsically safe database server, supporting regular automatic backup, and the modification needs to be verified by the administrator permission, to ensure the safety and traceability of the preset rules.

[0087] S702: triggering the corresponding level of warning signal according to the rock burst danger level, which includes the sound and light prompt signal of the local area for the first warning, the communication terminal pushing the warning information and accompanying sound and light alarm for the monitoring area for the second warning, and the emergency alarm signal of the whole monitoring network coverage area for the third warning, triggering the personnel evacuation notification at the same time.

[0088] Specifically, when S702 triggers the early warning signal, the specific implementation methods of each level of early warning signal are as follows: the first-level early warning deploys explosion-proof sound and light prompt devices in local areas such as tunnel intersections and working surfaces in the core monitoring area. The device adopts an explosion-proof shell, built-in yellow LED warning light (flashing frequency 2Hz) and 110dB buzzer (sounding at intervals of 3 seconds), and the prompt content is "Level 1 early warning: There is a potential impact risk in the local area, please strengthen monitoring"; the second-level early warning pushes text information (including early warning level, coordinates of the area involved and risk warnings) to the secondary monitoring area and underground workers within a radius of 500 meters through the mine intrinsically safe communication terminal (such as KT425 mine walkie-talkie), and starts the secondary monitoring at the same time. The red rotating warning light (rotating at 60 rpm) and the 130dB intermittent alarm sound (sounding for 1 second and paused for 1 second) in the area are turned on, and a warning pop-up window is popped up on the display of the ground monitoring center simultaneously; the third-level warning triggers the emergency alarm system in the entire monitoring network coverage area, including explosion-proof loudspeakers arranged every 100 meters underground (playing "Third-level warning: emergency evacuation, evacuate immediately along the safe passage" in a loop), and the sound and light alarm device of the ground dispatch center (red flashing light and 150dB alarm sound). At the same time, the underground personnel positioning system sends a mandatory pop-up notification to all personnel terminals in high-risk areas. The notification content includes the three nearest evacuation routes and the estimated evacuation time, and is linked with the mine emergency broadcast system to achieve full area coverage.

[0089] S703: Record the triggering time, level and coverage of the warning signal to form a warning log.

[0090] Specifically, when S703 records the warning log, the system automatically generates a structured log file, the specific contents of which include: warning trigger time (accurate to milliseconds), the corresponding rock burst hazard level (e.g., high-risk level), warning level (e.g., level 3 warning), coverage (accurate to the tunnel number and coordinate interval, e.g., "No. 3 coal seam transport tunnel K0+200 to K0+500 section"), and trigger basis (e.g., "LSTM model outputs a high-risk probability of 85%, and random forest model outputs a high-risk probability of 82%"). Logs are stored in a distributed database (e.g., HBase), each log is associated with a unique identifier (UUID), and is synchronously backed up to a remote server on the ground with a retention period of 3 years. At the same time, the system provides a log query interface that supports filtering by time, level, region, and other conditions. It can be exported in Excel or PDF format to facilitate subsequent warning effect evaluation and accident tracing analysis.

[0091] By presetting the corresponding relationship between the danger level and the early warning level, the judgment standard is clear, and the early warning normativity and reliability are improved; the hierarchical early warning signal (low-danger local sound and light, medium-danger regional push, high-danger whole-area alarm and evacuation linkage) is adopted to realize the accurate transmission of risk information, balance the production interference and emergency efficiency; the early warning log is recorded to form a closed loop, provide data support for evaluation and optimization, and overall improve the early warning scientificity, effectiveness and manageability to ensure the safety of the mine.

[0092] As shown in Figure 9 , step S80 can specifically include: S801: if the early warning signal is of high danger level, the spatial distribution density of the monitoring points in the monitoring network is increased and the data acquisition frequency is increased, at least one of support strength reinforcement, stress directional pressure relief and surrounding rock grouting reinforcement is selected to implement prevention and control, and a personnel emergency evacuation instruction is triggered.

[0093] Specifically, for the high danger level early warning signal, the monitoring network adjustment is as follows: the density of stress sensors in the core monitoring area is increased from 1 sensor per 5 meters to 1 sensor per 2 meters, the temperature and radon sensor spacing in the secondary monitoring area is reduced from 10 meters to 5 meters, and the number of sampling points of the peripheral laser displacement sensor array is increased by 50%; in terms of data acquisition frequency, the stress and strain data is increased from 1 Hz to 10 Hz, the microseismic and acoustic signal sampling rate is increased from 2 kHz to 10 kHz, and high-frequency dynamic changes are ensured to be captured. In the implementation of the prevention and control measures, the support strength reinforcement adopts Φ22mm high-strength anchor rod (original specification Φ20mm) arranged at an interval of 1.0m x 1.0m, and Φ17.8mm prestressed anchor cable (interval row spacing 2m x 2m) is additionally arranged; the stress directional pressure relief is constructed by pressure relief drillings with a diameter of 150mm and a depth of 15m, the hole spacing is 3m, and the drillings are arranged in a fan shape on the two sides of the roadway; the surrounding rock grouting reinforcement adopts cement-silicate double slurry (volume ratio 1:1), the grouting pressure is controlled at 3-5MPa, the grouting hole depth is 8m, and the spacing is 5m. The personnel emergency evacuation instruction sends a forced evacuation instruction to the personnel terminal in the high danger area through the underground personnel positioning system (KJ237 type), simultaneously starts the emergency broadcast system to circularly broadcast the evacuation route, and the dispatch center monitors the personnel evacuation progress in real time until all personnel leave the dangerous area.

[0094] S802: if the early warning signal is of medium danger level, the basic layout of the monitoring network is maintained and the data acquisition frequency is moderately increased, and at least one of support strength reinforcement or stress directional pressure relief is selected to implement prevention and control.

[0095] Specifically, S802 responds to the medium-risk level warning, the monitoring network maintains the basic layout of 5 meters per sensor in the core area and 10 meters per sensor in the secondary area, and the data collection frequency is moderately increased, among which the mechanical data is adjusted from 1 Hz to 5 Hz, and the fluctuation signal is adjusted from 2 kHz to 5 kHz. If the control measure is to strengthen the support strength, the Φ20 mm anchor is encrypted to 1.2 m x 1.2 m spacing, and a steel belt is added to the roadway roof. If stress directional pressure relief is selected, a pressure relief hole with a diameter of 120 mm and a depth of 10 m is constructed, with a hole spacing of 5 m, and a single row is arranged on one side of the roadway. If both are combined, half of the above parameters are implemented. During the implementation of the measures, a special person is arranged to record the support installation torque, pressure relief hole deslagging amount and other parameters every 2 hours to ensure that the construction quality meets the design requirements.

[0096] S803: If the warning signal is low-risk level, maintain the regular layout of the monitoring network and the data collection frequency, and select the surrounding rock grouting reinforcement or conventional support maintenance measures to implement control. Specifically, S803 for low-risk level warning, the monitoring network maintains the regular layout (5 meters per sensor in the core area, 10 meters per sensor in the secondary area, and 20 meters per laser displacement point in the peripheral area), and the data collection frequency remains the initial setting (1 Hz for mechanical data, 2 kHz for fluctuation data, and 5 minutes per time for environmental data). If the control measure is to use surrounding rock grouting reinforcement, low-concentration cement slurry (water-cement ratio 1:1.5) is selected, the grouting pressure is 1-2 MPa, the grouting hole depth is 5 m, and the spacing is 10 m. If conventional support maintenance is implemented, the anchor nut torque (ensure ≥300N m), anchor cable pretightening force (≥150 kN) is arranged for each shift inspection, loose parts are tightened in time, and deformed support components are replaced locally.

[0097] S804: Feedback the monitoring adjustment parameters and the implementation results of the control measures of each level to the monitoring network.

[0098] Specifically, the S804 feedback mechanism is as follows: the monitoring adjustment parameters (such as the sensor density and sampling frequency of the high-risk level), the implementation data (such as the number of anchor rods, the total length of pressure relief holes, and the grouting amount) and the implementation effect (such as the stress value change rate and the microseismic frequency attenuation rate) of the control measures are integrated into structured data, which is transmitted to the ground database through industrial Ethernet. The system automatically generates a "monitoring adjustment and control effect comparison table", which includes multi-dimensional monitoring data comparison curves (such as stress time history curve and microseismic energy distribution graph) before and after the implementation of the measures, and calculates the risk reduction rate of the control measures (formula: (risk probability before measures - risk probability after measures) / risk probability before measures x 100%). This feedback data is used to dynamically optimize the monitoring network layout algorithm and the prediction model parameters, for example, when the risk is significantly reduced after implementing stress pressure relief in a certain area, the system will automatically increase the weight of this area in the model training, improving the accuracy of subsequent warning and control.

[0099] For high-risk level early warning, the real-time monitoring accuracy of the dangerous area can be enhanced by increasing the monitoring point density and data collection frequency, combined with the combination of support strength reinforcement, stress directional pressure relief and surrounding rock grouting reinforcement control measures, which can quickly curb the risk escalation and protect the safety of personnel; for medium-risk level early warning, maintaining the basic monitoring layout and moderately increasing the collection frequency can ensure the effectiveness of monitoring while avoiding resource waste, and selecting at least one of support reinforcement or stress directional pressure relief can control the risk diffusion; for low-risk level early warning, maintaining regular monitoring and control measures can achieve normalized management of risks and reduce unnecessary cost investment; the mechanism of feeding back the adjustment parameters and execution results to the monitoring network can form a closed-loop optimization, continuously improving the adaptability of the monitoring system and the effectiveness of the control measures.

[0100] As shown in Figure 9 The embodiment provides a deep well rock burst early warning system based on multi-dimensional monitoring, which comprises: A data collection module, which comprises a monitoring network arranged in a deep well tunnel and surrounding rock, is used for collecting multi-dimensional monitoring data including stress value, strain value, microseismic event parameter, acoustic signal, temperature data, radon concentration value, electromagnetic radiation intensity and tunnel surface three-dimensional deformation data in real time; A space-time registration module, which is used for space-time registration of the multi-dimensional monitoring data, to obtain a multi-source monitoring data set aligned in space-time; A noise suppression module, which is used for dynamic noise suppression of the multi-source monitoring data set based on physical characteristics, to obtain a data set after noise suppression; A feature screening module, which is used for screening features related to rock burst intensity from the data set after noise suppression through a mutual information entropy algorithm, to obtain a data set after feature optimization; A feature extraction module, which is used for extracting frequency domain features, time domain features and time-frequency domain features from the data set after feature optimization through a principal component extraction method and a phase space reconstruction method, to form a multi-dimensional state space data set; A level prediction module, which comprises a pre-trained rock burst danger level prediction model, is used for inputting the multi-dimensional state space data set into the model to obtain a rock burst danger level; An early warning triggering module, which is used for triggering a corresponding level of early warning signal according to the rock burst danger level; A dynamic adjustment module, which is used for dynamically adjusting the layout of the monitoring network and control measures based on the corresponding level of early warning signal.

[0101] In further embodiments, the data collection module specifically comprises: The monitoring partition unit is configured to determine a core monitoring area, a secondary monitoring area and a peripheral monitoring area of the monitoring network based on geological mechanics parameters of a deep shaft tunnel and a rock burst historical risk zoning; The sensor configuration unit is configured to set a stress sensor, a strain sensor, a microseismic monitoring system and an acoustic wave monitoring device in the core monitoring area, set a temperature sensor, a radon concentration sensor and an electromagnetic radiation monitoring device in the secondary monitoring area, and set a laser displacement sensor array in the peripheral monitoring area; The parameter setting unit is configured to set acquisition parameters of the multi-dimensional monitoring data, including a sampling frequency, a trigger condition and a range of a quantity; The transmission link unit is configured to build a hierarchical data transmission link based on importance levels of the monitoring areas, wherein data of the core monitoring area is transmitted through an industrial Ethernet channel with a redundant backup, data of the secondary monitoring area is transmitted through a mine Ethernet channel, and data of the peripheral monitoring area is transmitted through a wireless ad hoc network, so as to realize real-time transmission of the multi-dimensional monitoring data; The space-time labeling unit is configured to perform time stamp synchronization and geological coordinate labeling on the multi-dimensional monitoring data.

[0102] In further embodiments, the space-time registration module specifically includes: The coordinate system establishment unit is configured to establish a space-time reference coordinate system based on a three-dimensional geological model of a deep shaft tunnel, the space-time reference coordinate system taking a starting point of the tunnel as an origin, a tunnel axial direction as an X axis, a direction perpendicular to a tunnel trend as a Y axis, a direction perpendicular to a tunnel floor as a Z axis, and being associated with an absolute geological coordinate; The time calibration unit is configured to perform time stamp calibration on the multi-dimensional monitoring data, unify collection times of various monitoring devices to a standard time axis of the space-time reference coordinate system, and perform time dimension alignment on time asynchronous data by using a linear interpolation method; The space conversion unit is configured to build a space coordinate conversion matrix of a setting position of each monitoring device based on the space-time reference coordinate system, and convert original space data of different monitoring devices to three-dimensional coordinates under the space-time reference coordinate system; The space gridding unit is configured to perform space gridding processing on monitoring data with uneven spatial distribution by using a Kriging interpolation method, to form a spatially continuous monitoring data field, and the monitoring data field is used to integrate spatial correlation relationships of the multi-dimensional monitoring data; An error checking unit is configured to calculate a registration error according to a time synchronization error and a spatial coordinate deviation, wherein the time synchronization error is a deviation value of a time stamp of the monitoring data after calibration from a standard time axis, and the spatial coordinate deviation is a deviation amount of the three-dimensional coordinate after conversion from a theoretical coordinate of the space-time reference coordinate system; when the registration error exceeds a registration accuracy threshold, re-execution of time calibration or spatial conversion is triggered; and when the registration error meets a requirement of the registration accuracy threshold, the multi-source monitoring data set in space-time alignment is output.

[0103] In further embodiments, the noise suppression module specifically includes: A noise identification unit is configured to classify and identify noise types corresponding to each dimension of data based on physical properties of the multi-source monitoring data, wherein the mechanical data includes stress values and strain values, the fluctuation signals include microseismic event parameters and acoustic signals, the environmental parameters include temperature data and radon concentration values, and the electromagnetic and optical data includes electromagnetic radiation intensity and three-dimensional deformation data of a roadway surface; An algorithm matching unit is configured to use a noise reduction algorithm matched with physical properties of the data to suppress noise according to the identified noise types, which includes using a drift compensation algorithm based on physical change trends for the mechanical data, using an adaptive filtering algorithm for the fluctuation signals, using a baseline correction algorithm for the environmental parameters, and using an interference subspace separation algorithm for the electromagnetic and optical data; A parameter adjustment unit is configured to monitor a data variability in the noise suppression process in real time, and use the data variability as a feedback to dynamically adjust parameters of the noise reduction algorithm, so that the suppression intensity is adapted to dynamic changes of the noise; A threshold checking unit is configured to calculate a signal-to-noise ratio and a feature retention degree of the data after noise suppression, and output the data set after noise suppression when the signal-to-noise ratio reaches a signal-to-noise ratio threshold and the feature retention degree reaches a feature integrity threshold; otherwise, re-adjustment of the parameters of the noise reduction algorithm is triggered.

[0104] In further embodiments, the feature screening module specifically includes: A model construction unit is configured to construct a mutual information entropy calculation model between each feature in the data set after noise suppression and rock burst danger, wherein the mutual information entropy calculation model is based on a mutual information entropy formula in information theory, and quantifies an information correlation degree between a feature variable and a rock burst danger level; An entropy value calculation unit is configured to calculate mutual information entropy values of each feature through the mutual information entropy calculation model, wherein the mutual information entropy values represent a prediction contribution degree of the feature to the rock burst danger; A feature screening unit is configured to screen features with mutual information entropy values greater than a feature contribution degree threshold according to the calculated mutual information entropy values, to form a feature subset related to rock burst intensity; A redundancy elimination unit is configured to perform a redundancy analysis on the feature subset, eliminate redundant features with similar mutual information entropy values and a correlation higher than a feature redundancy threshold, and obtain a feature-optimized dataset.

[0105] In further embodiments, the feature extraction module specifically includes: A principal component extraction unit is configured to construct a covariance matrix based on the feature-optimized dataset, perform eigenvalue decomposition on the covariance matrix, extract principal components with a variance contribution rate exceeding a preset threshold, and form a dimension-reduced feature subset. A phase space construction unit is configured to determine embedding dimension and time delay parameters based on an autocorrelation function and a mutual information function of the dimension-reduced feature subset, and construct a high-dimensional phase space based on a phase space reconstruction method. A feature extraction unit is configured to extract frequency domain features, time domain features, and time-frequency domain features of each feature in the high-dimensional phase space, respectively, wherein the frequency domain features are obtained by Fourier transform, the time domain features are calculated based on time series statistical characteristics, and the time-frequency domain features are extracted by wavelet transform. A feature fusion unit is configured to perform feature fusion on the extracted frequency domain features, time domain features, and time-frequency domain features, and form a multi-dimensional state space dataset.

[0106] In further embodiments, the pre-trained rock burst danger level prediction model in the level prediction module is constructed using a long short-term memory neural network combined with a random forest algorithm optimized by a genetic algorithm.

[0107] In further embodiments, the early warning triggering module specifically includes: A level correspondence unit is configured to preset a correspondence between rock burst danger levels and early warning levels, wherein a low danger level corresponds to a first-level early warning, a medium danger level corresponds to a second-level early warning, and a high danger level corresponds to a third-level early warning. A signal triggering unit is configured to trigger an early warning signal of a corresponding level according to the rock burst danger level, which includes a local area sound-light prompt signal for a first-level early warning, a communication terminal push early warning information accompanied by a sound-light alarm for a monitoring area for a second-level early warning, and an emergency alarm signal for a full monitoring network coverage area for a third-level early warning, while triggering a personnel evacuation instruction. A log recording unit is configured to record the triggering time, level, and coverage range of the early warning signal, and form an early warning log.

[0108] In further embodiments, the dynamic adjustment module specifically includes: A high danger adjustment unit is configured to increase the spatial distribution density of monitoring points in the monitoring network and increase the data acquisition frequency when the early warning signal is at a high danger level, select a combination of support strength reinforcement, stress directional pressure relief, and surrounding rock grouting reinforcement for implementation of prevention and control, and trigger an emergency evacuation instruction for personnel. The medium-risk adjustment unit is used to maintain the basic layout of the monitoring network and appropriately increase the data collection frequency when the early warning signal is at the medium-risk level, and select at least one of support strength enhancement or stress-oriented pressure relief to implement prevention and control; The low-risk adjustment unit is used to maintain the normal layout of the monitoring network and the frequency of data collection when the early warning signal is at a low-risk level, and to select surrounding rock grouting reinforcement or conventional support and maintenance measures for prevention and control; The feedback unit is used to feed back the monitoring adjustment parameters and prevention and control measures execution results corresponding to each level to the monitoring network.

[0109] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements any of the above methods when the program is executed by a processor. The present invention further provides an electronic device. The electronic device of an embodiment of the present invention includes: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement a deep well rock burst warning method based on multi-dimensional monitoring provided by the present invention. Figure 10 As shown, the computer system 800 of the electronic device includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 802 or programs loaded from a storage unit 808 into a random access memory (RAM) 803. RAM 803 also stores various programs and data required for the operation of computer system 800. CPU 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to bus 804. The following components are connected to I / O interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 808 including a hard disk; and a communication section 809 including a network interface card such as a LAN card or modem. Communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like, is mounted on the drive 810 as needed, so that a computer program read therefrom is installed into the storage section 808 as needed.

[0110] The above specific implementations do not limit the scope of protection of the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A deep well rock burst early warning method based on multi-dimensional monitoring, characterized in that: The following steps are involved: S10: Utilize the monitoring network deployed in deep mine tunnels and surrounding rocks to collect multi-dimensional monitoring data in real time, including stress values, strain values, microseismic event parameters, acoustic wave signals, temperature data, radon gas concentration values, electromagnetic radiation intensity, and three-dimensional deformation data of the tunnel surface; S20: performing spatiotemporal registration on the multi-dimensional monitoring data to obtain a spatiotemporally aligned multi-source monitoring data set; S30: performing dynamic noise suppression based on physical characteristics on the multi-source monitoring dataset to obtain a noise-suppressed dataset; S40: filtering features related to rock burst pressure from the noise-suppressed data set using a mutual information entropy algorithm to obtain a feature-optimized data set; S50: extracting frequency domain features, time domain features, and time-frequency domain features from the feature-optimized data set through a principal component extraction method and a phase space reconstruction method to form a multidimensional state space data set; S60: Inputting the multidimensional state space data set into a pre-trained rock burst hazard level prediction model to obtain the rock burst hazard level; S70: triggering a warning signal of a corresponding level according to the rock burst hazard level; S80: Based on the early warning signal of the corresponding level, dynamically adjust the layout and prevention and control measures of the monitoring network.

2. The method according to claim 1, characterized in that Step S10 specifically includes: S101: Based on the geomechanical parameters of deep well tunnels and the historical risk zoning of rock burst, determine the core monitoring area, secondary monitoring area and peripheral monitoring area of ​​the monitoring network; S102: Installing stress sensors, strain sensors, a microseismic monitoring system, and an acoustic wave monitoring device in the core monitoring area; installing temperature sensors, radon gas concentration sensors, and an electromagnetic radiation monitoring device in the secondary monitoring area; and installing a laser displacement sensor array in the peripheral monitoring area; S103: Setting the acquisition parameters of the multi-dimensional monitoring data, including sampling frequency, trigger conditions and range; S104: Constructing a hierarchical data transmission link based on the importance level of the monitoring area, wherein the data of the core monitoring area is transmitted via an industrial Ethernet channel with redundant backup, the data of the secondary monitoring area is transmitted via a mining Ethernet channel, and the data of the peripheral monitoring area is aggregated and transmitted via a wireless ad hoc network, thereby realizing real-time transmission of the multi-dimensional monitoring data; S105: Perform time stamp synchronization and geological coordinate marking on the multi-dimensional monitoring data.

3. The method according to claim 1, characterized in that Step S20 specifically includes: S201: establishing a spatiotemporal reference coordinate system based on a three-dimensional geological model of a deep well tunnel, wherein the spatiotemporal reference coordinate system has the tunnel starting point as the origin, the tunnel axis as the X-axis, the vertical tunnel direction as the Y-axis, and the vertical tunnel floor as the Z-axis, and is associated with absolute geological coordinates; S202: performing time stamp calibration on the multi-dimensional monitoring data, unifying the acquisition time of various monitoring devices used to acquire the multi-dimensional monitoring data to the standard time axis of the spatiotemporal reference coordinate system, and aligning the time dimension of the time-asynchronous data using linear interpolation; S203: constructing a spatial coordinate conversion matrix for each monitoring device location based on the time-space reference coordinate system, and converting the original spatial data of different monitoring devices into three-dimensional coordinates in the time-space reference coordinate system; S204: Using the Kriging interpolation method to perform spatial gridding processing on the spatially unevenly distributed monitoring data to form a spatially continuous monitoring data field, wherein the monitoring data field is used to integrate the spatial correlation relationship of the multi-dimensional monitoring data; S205: Calculate the registration error based on the time synchronization error and the spatial coordinate deviation, where the time synchronization error is the deviation value between the timestamp of the calibrated monitoring data and the standard time axis, and the spatial coordinate deviation is the deviation between the converted three-dimensional coordinates and the theoretical coordinates of the spatiotemporal reference coordinate system; when the registration error exceeds the registration accuracy threshold, return to step S202 to re-perform time calibration or space conversion; when the registration error meets the requirements of the registration accuracy threshold, output the spatiotemporally aligned multi-source monitoring data set.

4. The method according to claim 1, wherein Step S30 specifically includes: S301: Based on the physical properties of multi-source monitoring data, classify and identify the noise type corresponding to each dimension of data. Mechanical data includes stress and strain values, wave signals include microseismic event parameters and acoustic wave signals, environmental parameters include temperature data and radon gas concentration values, and electromagnetic and optical data include electromagnetic radiation intensity and three-dimensional deformation data of the tunnel surface. S302: Based on the identified noise type, noise reduction algorithms matching the physical characteristics of the data are used to suppress the noise, including a drift compensation algorithm based on physical change trends for the mechanical data, an adaptive filtering algorithm for the fluctuation signal, a baseline correction algorithm for the environmental parameters, and an interference subspace separation algorithm for the electromagnetic and optical data. S303: real-time monitoring of data variability during the noise suppression process, and using the data variability as feedback to dynamically adjust parameters of the noise reduction algorithm to adapt the suppression strength to dynamic changes in noise; S304: Calculate the signal-to-noise ratio and feature retention of the noise-suppressed data. When the signal-to-noise ratio reaches the signal-to-noise ratio threshold and the feature retention reaches the feature integrity threshold, output the noise-suppressed data set; otherwise, return to step S302 to readjust the parameters of the noise reduction algorithm.

5. The method according to claim 1, wherein Step S40 specifically includes: S401: constructing a mutual information entropy calculation model between each feature in the noise-suppressed dataset and the rock burst hazard, wherein the mutual information entropy calculation model is based on the mutual information entropy formula in information theory and quantifies the degree of information correlation between the feature variable and the rock burst hazard level; S402: Calculating the mutual information entropy value of each feature using the mutual information entropy calculation model, where the mutual information entropy value represents the contribution of the feature to the prediction of rock burst hazard; S403: Based on the calculated mutual information entropy value, select features whose mutual information entropy value is greater than a feature contribution threshold to form a feature subset related to rock burst pressure; S404: performing redundancy analysis on the feature subset, eliminating redundant features with similar mutual information entropy values ​​and correlations higher than a feature redundancy threshold, and obtaining a feature-optimized data set.

6. The method according to claim 1, characterized in that Step S50 specifically includes: S501: constructing a covariance matrix for the feature-optimized data set, performing eigenvalue decomposition on the covariance matrix, extracting principal components whose variance contribution rate exceeds a preset threshold, and forming a feature subset after dimensionality reduction; S502: Based on a phase space reconstruction method, determining an embedding dimension and a time delay parameter according to the autocorrelation function and the mutual information function of the feature subset after dimensionality reduction, and constructing a high-dimensional phase space; S503: Extracting frequency domain features, time domain features, and time-frequency domain features of each feature in the high-dimensional phase space, wherein the frequency domain features are obtained by Fourier transform, the time domain features are calculated based on statistical characteristics of time series, and the time-frequency domain features are extracted by wavelet transform; S504: Fusing the extracted frequency domain features, time domain features, and time-frequency domain features to form a multi-dimensional state space data set.

7. The method according to claim 1, characterized in that The pre-trained rock burst hazard level prediction model is constructed by using a long short-term memory neural network combined with a random forest algorithm optimized by a genetic algorithm.

8. The method according to claim 1, characterized in that Step S70 specifically includes: S701: Preset the correspondence between rock burst hazard level and warning level, where low hazard level corresponds to level 1 warning, medium hazard level corresponds to level 2 warning, and high hazard level corresponds to level 3 warning; S702: triggering a warning signal of a corresponding level according to the rock burst hazard level, including a level 1 warning signal of an audible and visual prompt signal in a local area, a level 2 warning signal of a warning message pushed to a communication terminal in the monitoring area accompanied by an audible and visual alarm, and a level 3 warning signal of an emergency alarm signal for the entire monitoring network coverage area, which also triggers a personnel evacuation notice; S703: Record the triggering time, level and coverage of the warning signal to form a warning log.

9. The method according to claim 8, characterized in that Step S80 specifically includes: S801: If the early warning signal is high-risk, increase the spatial distribution density of monitoring points in the monitoring network and the frequency of data collection. A combination of support strength enhancement, stress-oriented unloading, and surrounding rock grouting reinforcement is selected for prevention and control, and an emergency evacuation order is triggered. S802: If the warning signal is a medium-risk level, maintain the basic layout of the monitoring network and appropriately increase the frequency of data collection. Select at least one of support strength enhancement or stress-oriented unloading for prevention and control. S803: If the warning signal is low-risk, maintain the normal layout of the monitoring network and the frequency of data collection, and select surrounding rock grouting reinforcement or conventional support and maintenance measures for prevention and control; S804: Feedback the monitoring adjustment parameters and prevention and control measures execution results corresponding to each level to the monitoring network.

10. A deep well rock burst early warning system based on multi-dimensional monitoring, characterized in that: include: A data acquisition module, comprising a monitoring network arranged in the deep well tunnel and surrounding rock, for real-time acquisition of multi-dimensional monitoring data including stress values, strain values, microseismic event parameters, acoustic wave signals, temperature data, radon gas concentration values, electromagnetic radiation intensity, and three-dimensional deformation data of the tunnel surface; A spatiotemporal registration module, configured to perform spatiotemporal registration on the multi-dimensional monitoring data to obtain a spatiotemporally aligned multi-source monitoring data set; a noise suppression module, configured to perform dynamic noise suppression based on physical characteristics on the multi-source monitoring dataset to obtain a noise-suppressed dataset; a feature screening module, configured to screen features related to rock burst pressure from the noise-suppressed data set using a mutual information entropy algorithm to obtain a feature-optimized data set; A feature extraction module is used to extract frequency domain features, time domain features and time-frequency domain features from the feature-optimized data set through a principal component extraction method and a phase space reconstruction method to form a multidimensional state space data set; A level prediction module, the level prediction module including a pre-trained rock burst hazard level prediction model, for inputting the multi-dimensional state space data set into the model to obtain the rock burst hazard level; An early warning trigger module is used to trigger an early warning signal of a corresponding level according to the rock burst hazard level; A dynamic adjustment module is used to dynamically adjust the layout and prevention and control measures of the monitoring network based on the early warning signals of the corresponding levels.

Images