Early identification method and device for rockburst based on physical information injection and medium

Abstract

The invention discloses a rockburst early recognition method and device based on physical information injection and a medium, and belongs to the technical field of underground engineering disaster monitoring and intelligent early warning. The method comprises the following steps: acquiring acoustic emission, microseismic, stress and displacement data, and performing space-time alignment in combination with lithology, joints and other static geological attributes to generate a high-dimensional feature cube; cubic input physical information is injected into a converter for modeling, and intermediate features such as energy accumulation rate and three-dimensional danger confidence distribution are output; retrieving a rockburst case library based on the intermediate features, and fusing similar cases to generate an induction mechanism interpretation and prevention and control suggestion text; a potential fracture surface position is extracted for visualization; triggering an edge alarm based on the risk level and recording execution feedback; input distribution and physical residual distribution are monitored, and increment fine tuning is triggered according to the drift state and feedback. According to the method, the technical effects of reducing the false report and missing report rate and improving the early warning timeliness and the spatial positioning precision are achieved.

Description

Technical Field

[0001] The present application relates to the technical field of underground engineering disaster monitoring and intelligent early warning, and in particular to a method, device and medium for early identification of rockburst based on physical information injection. Background Art

[0002] Rockbursts (also known as rock bursts) are common and devastating dynamic hazards in underground tunnels, such as deep coal and metal mines. They manifest as sudden instability of the rock mass under stress, instantly releasing enormous elastic energy. This can easily lead to damage to tunnel support structures, mining equipment, and even casualties. Currently, rockburst monitoring and early warning rely primarily on various sensors installed around the tunnels and working faces. These include triaxial strain gauges for monitoring surrounding rock stress, acoustic emission (AE) sensors for capturing elastic wave signals generated by rock fracture, and microseismic monitoring systems for locating larger-scale rock failure events. Based on the data collected by these sensors, existing technologies generally employ threshold-based alarm triggering. This means that an alarm is issued when the monitored stress value, AE energy, or microseismic event intensity exceeds a preset safety threshold. Additionally, some methods attempt to apply traditional machine learning algorithms (such as support vector machines and random forests) to statistically analyze monitoring data sequences in an effort to identify anomalous patterns. These technologies have, to a certain extent, provided assurance for underground engineering safety and formed the foundation for technological development in this field.

[0003] However, these existing technologies suffer from significant shortcomings. First, alarm methods that rely on fixed thresholds are ineffective in practical applications. Due to the complex and variable geological conditions of deep rock formations, significant lithologic differences (such as hard rock, soft rock, and interbedded waste), varying degrees of structural joint development, and dynamic shifts in stress fields caused by mining activities, uniform thresholds are difficult to adapt to all operating conditions. This leads to high rates of false alarms (identifying non-hazardous signals as hazardous) and false alarms (failing to identify actual hazards), making the accuracy and reliability of early warnings difficult to meet high safety requirements. Second, existing machine learning methods typically treat monitoring data as purely statistical sequences and fail to effectively incorporate the physical mechanisms of rock failure. These models ignore the rock's constitutive relationships (stress-strain laws), the tensor distribution characteristics of the in-situ stress field, and the key physical processes of energy accumulation and dissipation. This results in poor generalization when faced with complex geological conditions (such as high burial depth, strong tectonic stresses, and areas of lithologic abrupt changes). The prediction results lack physical plausibility and are difficult to explain. Furthermore, with the advancement of mine digitalization, multi-source heterogeneous data such as acoustic emission, microseismicity, stress, and displacement can be collected in real time. However, existing technologies lack the ability to efficiently integrate and deeply couple these multi-dimensional, spatiotemporal-related data, and fail to fully utilize the overall value of the data for early and accurate risk identification and spatial positioning. The timeliness and spatial accuracy of early warnings need to be improved.

[0004] Therefore, how to reduce the false alarm and missed alarm rates and improve the timeliness of warnings and spatial positioning accuracy has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The embodiments of the present application provide a method, device, and medium for early identification of rockburst based on physical information injection, which are used to solve the following technical problems: how to reduce the false alarm and missed alarm rate, and improve the timeliness of warning and spatial positioning accuracy.

[0006] In a first aspect, an embodiment of the present application provides a method for early identification of rockbursts based on physical information injection, the method comprising: collecting acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data in an underground engineering area; combining the collected acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data with static geological attribute data, performing time synchronization and spatial coordinate alignment to generate high-dimensional feature cube data with a unified time and space reference; wherein the static geological attribute data includes lithology, structural joints and strike and dip information; inputting the high-dimensional feature cube data into a preset physical information injection converter for modeling to obtain intermediate feature vector data including energy accumulation rate, stress concentration factor and fracture evolution potential, as well as three-dimensional hazard confidence distribution data; searching a preset rockburst case knowledge base based on the intermediate feature vector data to obtain similar historical case data, The similar historical case data is spliced ​​and fused with the intermediate feature vector data to generate natural language text data containing explanations of potential inducing mechanisms and suggestions for prevention and control measures; three-dimensional risk heat map data is generated based on the three-dimensional hazard confidence distribution data, potential fracture surface spatial position data is extracted based on the three-dimensional risk heat map data, and the three-dimensional risk heat map data, potential fracture surface spatial position data and the natural language text data are visualized; risk assessment is performed based on the three-dimensional hazard confidence distribution data, and an edge alarm signal is triggered when the risk level reaches a preset threshold, and the on-site execution feedback data for the prevention and control measures is recorded; the input distribution characteristics of the high-dimensional feature cube data and the physical residual distribution characteristics generated by the modeling processing are monitored, the model drift state is judged based on the monitoring results, and the incremental fine-tuning process of the model is triggered according to the model drift state and the execution feedback data.

[0007] In one implementation of the present application, the collected acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data are combined with static geological attribute data for time synchronization and spatial coordinate alignment to generate high-dimensional feature cube data with a unified time and space reference, specifically including: based on the preset IEEE-1588PTP precise time protocol, the sensor nodes that collect the acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data are time synchronized to obtain synchronized timestamp data; based on the preset RTK-SLAM algorithm and laser scanning point cloud data, the three-dimensional spatial coordinate frame data of the underground engineering area is established; the acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data are synchronized to generate high-dimensional feature cube data with a unified time and space reference. The deformation and displacement data are subjected to spatiotemporal mapping based on the synchronous timestamp data and the three-dimensional spatial coordinate frame data to obtain spatially aligned monitoring point data; sliding window filtering, noise suppression, and energy envelope quantization processing are performed on the acoustic emission waveform data and microseismic waveform data at the edge computing node to obtain preprocessed waveform feature data; adaptive Kalman filtering is performed on the triaxial stress data and tunnel deformation and displacement data to suppress noise and compensate for missing values ​​by linear interpolation to obtain preprocessed stress and displacement feature data; the preprocessed waveform feature data, the preprocessed stress and displacement feature data, and the static geological attribute data are organized into high-dimensional feature cube data indexed by time dimension, space dimension, and feature dimension according to a preset time window and spatial resolution.

[0008] In one implementation of the present application, the high-dimensional feature cube data is input into a preset physical information injection transformer for modeling to obtain intermediate feature vector data including energy accumulation rate, stress concentration factor and crack evolution potential, as well as three-dimensional hazard confidence distribution data, specifically including: performing frequency domain transformation processing on the high-dimensional feature cube data to obtain frequency domain feature data; inputting the frequency domain feature data into the multi-head self-attention layer of the physical information injection transformer; when each attention head of the multi-head self-attention layer calculates the similarity between the query vector and the key vector, explicitly introducing the rock mass elastic-plastic constitutive residual data and the in-situ ground stress tensor data as penalty terms , calculate and obtain attention weight data that meets the mechanical constraints; wherein, the rock mass elastic-plastic constitutive residual data is obtained by calculating the difference between the predicted stress and the actual stress in real time based on the constitutive equation obtained by field measurement or experiment; based on the calculated attention weight data, the value vector is weightedly aggregated to obtain deep feature coupling data; the deep feature coupling data is input into the multi-expert gated subnetwork of the physical information injection converter; based on the multi-expert gated subnetwork and the lithologic attributes of the input data, the deep feature coupling data is dynamically routed to the corresponding expert subnetwork for processing to obtain the intermediate feature vector data and the three-dimensional hazard confidence distribution data.

[0009] In one implementation of the present application, based on the lithologic attributes of the input data, the multi-expert gating subnetwork dynamically routes the deep feature coupling data to the corresponding expert subnetwork for processing to obtain the intermediate feature vector data and the three-dimensional hazard confidence distribution data, specifically including: inputting the deep feature coupling data and the lithologic label data in the static geological attribute data into the gating network of the multi-expert gating subnetwork; learning the coupling pattern of the deep feature coupling data with different lithologic types based on the gating network to generate expert selection weight data; based on the expert selection weight data, dynamically routing the deep feature coupling data to one or more best-matched expert subnetworks; processing the routed deep feature coupling data based on the selected expert subnetwork, and outputting the intermediate feature vector data and the three-dimensional hazard confidence distribution data.

[0010] In one implementation of the present application, a preset rock burst case knowledge base is retrieved based on the intermediate feature vector data to obtain similar historical case data, and the similar historical case data is spliced ​​and fused with the intermediate feature vector data to generate natural language text data containing potential inducing mechanism explanations and prevention and control measures suggestions, specifically including: performing a first-stage coarse-grained vector similarity search in the rock burst case knowledge base based on the intermediate feature vector data to obtain a preliminary matching candidate case data set; performing a second-stage fine-grained search on the preliminary matching candidate case data set, the fine-grained search is based on static attribute vector similarity. A hierarchical matching calculation is performed on the similarity of the static attribute vector and the dynamic sequence vector to screen out target case data with high similarity; wherein, the static attribute vector includes lithology, strike, and joint density, and the dynamic sequence vector includes an energy attenuation curve and a fracture evolution trajectory; the feature representation of the screened target case data is spliced ​​with the intermediate feature vector data to form knowledge-enhanced feature input data; the knowledge-enhanced feature input data is input into a language macromodel that has been fine-tuned by instructions; the knowledge-enhanced feature input data is processed based on the language macromodel to automatically generate the natural language text data containing the potential inducing mechanism explanation and prevention and control measures suggestions.

[0011] In one implementation of the present application, three-dimensional risk heat map data is generated based on the three-dimensional hazard confidence distribution data, potential rupture surface spatial position data is extracted based on the three-dimensional risk heat map data, and the three-dimensional risk heat map data, potential rupture surface spatial position data and the natural language text data are visualized, specifically including: mapping the three-dimensional hazard confidence distribution data to the spatial voxels corresponding to the three-dimensional tunnel geographic information model to generate initial three-dimensional risk heat map data; performing a clustering analysis algorithm based on density and spatial connectivity on the initial three-dimensional risk heat map data to identify high-risk connected area data; calculating the minimum convex hull geometry data of each identified high-risk connected area data, the minimum convex hull geometry data representing the potential rupture surface spatial position data; rendering and displaying the three-dimensional tunnel scene superimposed with the initial three-dimensional risk heat map data, the convex hull boundary of the high-risk connected area data and the natural language text data on a three-dimensional visualization platform.

[0012] In one implementation of the present application, a risk assessment is performed based on the three-dimensional hazard confidence distribution data, and an edge alarm signal is triggered when the risk level reaches a preset threshold, and the execution feedback data of the prevention and control measures recommended on site is recorded, specifically including: real-time assessment of the risk level of the three-dimensional hazard confidence distribution data in a continuous time window; when it is detected that the risk level of a predetermined number of consecutive time windows reaches or exceeds the preset risk threshold level, it is determined that the alarm triggering condition is met; when the alarm triggering condition is met, the edge alarm signal is immediately sent to the edge alarm device deployed at the tunnel site to trigger an audible and visual alarm; an alarm information window is automatically popped up on the central dispatching platform, and the alarm information window includes a high-risk area identifier, the prevention and control measures recommended and key explanations in the natural language text data; the operation type, operation time and post-operation regional stability observation result data performed by on-site personnel according to the prevention and control measures recommended are collected and recorded to form the execution feedback data.

[0013] In one implementation of the present application, the input distribution characteristics of the high-dimensional feature cube data and the physical residual distribution characteristics generated by the modeling process are monitored, the model drift state is judged based on the monitoring results, and the incremental fine-tuning process of the model is triggered according to the model drift state and the execution feedback data, specifically including: real-time calculation of first distribution difference measurement data between the current input distribution characteristics of the high-dimensional feature cube data and the historical benchmark input distribution characteristics; real-time calculation of second distribution difference measurement data between the current physical residual distribution characteristics generated by the modeling process and the expected physical residual distribution characteristics; based on the first distribution difference measurement data and the second distribution difference measurement data, calculating comprehensive drift index data; when the comprehensive drift index data continuously exceeds the preset drift threshold for a predetermined time, it is determined that model drift has occurred, triggering a model degradation operation, automatically switching to a historical stable model version for inference, and generating model aging alarm information; collecting the execution feedback data, model prediction error data and actual alarm event data, and annotating them to form an incremental training sample data set; based on the incremental training sample data set, when a model performance degradation is detected or a regular update instruction is received, triggering the incremental fine-tuning training process of the physical information injection transformer to update the model parameters.

[0014] In a second aspect, an embodiment of the present application further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for early identification of rockbursts based on physical information injection as described in any one of the above contents.

[0015] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for early identification of rockburst based on physical information injection as described in any one of the above contents.

[0016] The embodiments of the present application provide a method, device, and medium for early identification of rockburst based on physical information injection, which have at least the following technical effects: By embedding the rock mass elastic-plastic constitutive residuals and the in-situ in-situ stress tensor as explicit priors in the self-attention weight calculation, the model is forced to adhere to rock mass mechanics conservation and volume compatibility constraints, fundamentally avoiding misinterpretation of non-hazardous signals such as environmental noise and local stress fluctuations as signs of rockburst. Furthermore, a multi-expert gating mechanism dynamically adapts to the heterogeneous response characteristics of complex rock types, such as hard rock, soft rock, and interbedded waste, significantly reducing the risk of missed reports due to varying geological conditions. It also maintains high generalization capabilities even under extreme conditions such as deep burial, high pressure, and developed fractures.

[0017] Based on a geological knowledge alignment mechanism called Retrieval Augmented Generation (RAG), real-time monitoring features are semantically matched against a historical rockburst case database at multiple levels, automatically generating a natural language explanation encompassing the "triggering mechanism - prevention and control measures" framework. This design transforms abstract numerical warnings into actionable engineering decision-making frameworks (e.g., "reinforcement anchor cable + directional pressure relief drilling parameters"). This allows dispatchers to simultaneously grasp the complete logical chain of "where the risk is high, why it's high, and how to address it," significantly improving the efficiency of on-site early warning information adoption and the accuracy of on-site response.

[0018] Leveraging sub-millisecond time synchronization and centimeter-level spatial registration of multi-source data, including acoustic emission and microseismic data, in a unified three-dimensional coordinate system, combined with the high-resolution hazard confidence field output by the physical information injection model, the system achieves millimeter-level spatial localization of potential rupture surfaces. The system automatically extracts connected blocks of the risk heat volume and calculates the minimum convex hull, providing intuitive spatial coordinate guidance for advanced support deployment and directional pressure relief drilling. This upgrades the traditional "regional-level" response of threshold alarms to "point-level" proactive prevention and control, significantly shortening the time window for disaster response.

[0019] By continuously monitoring the drift of input data distribution and physical residual distribution, combined with a closed-loop incremental learning mechanism based on on-site execution feedback, the model can dynamically adapt to changes in operating conditions, such as stress field evolution and rock structure variations caused by mining advancement. While protecting data privacy, the federated learning framework enables the collaborative accumulation of heterogeneous geological experience across mining areas, ensuring that the system can continue to optimize early warning performance even after large-scale deployment, essentially extending the technology's lifecycle.

[0020] High-reliability early warnings significantly reduce the frequency of manual inspections and production losses caused by accidental mining stoppages; explainable prevention and control suggestions shorten the plan formulation cycle and reduce the safety risks of experts frequently going deep into high-risk tunnels; the edge's minute-level response capability avoids emergency repair investments caused by the expansion of disasters, achieving an overall reduction in mine safety operation and maintenance costs from the three dimensions of early warning accuracy, manpower allocation, and disaster response. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flow chart of a method for early identification of rockburst based on physical information injection provided in an embodiment of the present application; Figure 2 A schematic diagram of the internal structure of a rockburst early identification device based on physical information injection provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0023] The embodiments of the present application provide a method, device, and medium for early identification of rockburst based on physical information injection, which are used to solve the following technical problems: how to reduce the false alarm and missed alarm rate, and improve the timeliness of warning and spatial positioning accuracy.

[0024] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0025] Figure 1 This is a flow chart for early identification of rockburst based on physical information injection provided in the embodiment of the present application. Figure 1 As shown, an embodiment of the present application provides a method for early identification of rockburst based on physical information injection, which specifically includes the following steps: Step 1: Collect acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation and displacement data within the underground engineering area.

[0026] Step 1.1: Acoustic emission waveform data acquisition. Transmitted waveform data: This refers to the elastic wave signals released by rock microfractures, used to detect early signs of fine cracks within the rock mass. High-frequency sensors (such as the PAC-SAMOS system) capture these signals at a sampling rate of 1 MHz or higher and a dynamic range of 100 dB, ensuring high sensitivity and interference resistance.

[0027] In a specific example, Mine A deployed PAC-SAMOS acoustic emission sensors every 50 meters in the E1203 working face tunnel roof. The sensors were positioned to avoid known fracture zones and focus on intact rock formations to capture microfracture events. After deployment, the system operated at a 1.2MHz sampling rate with a dynamic range of 100dB, ensuring the capture of elastic waves released by microcracks as small as 0.1mm. Real-time data was transmitted via explosion-proof armored cables to prevent signal integrity from being affected by underground gas or dust. This deployment successfully captured multiple microfracture events within 24 hours, providing initial input for the system.

[0028] Step 1.2: Acquisition of microseismic waveform data. Microseismic waveform data: refers to the vibration signal generated by macroscopic rock fracture events, monitored by broadband microseismic stations, with a frequency response range of 10Hz–1kHz and a sensitivity of 0.1m / s 2 , used to identify larger-scale rock mass instability.

[0029] In a specific example, a broadband microseismic station was deployed at the intersection of the tunnels 300m away from the working face in Mine A. The station spacing was set at 250m to cover the entire high-risk area. Sensitivity calibration (0.1m / s 2 ), with a fixed frequency response range of 10Hz–1kHz to accommodate deep, high-pressure rock formations. After deployment, the system successfully detected an energy release event (e.g., rock displacement), with real-time data uploaded via the industrial ring network. This implementation focuses on optimizing station spacing to ensure blind spot monitoring and provide reliable input for subsequent analysis.

[0030] Step 1.3: Triaxial stress data collection. Triaxial stress data: refers to the measured values ​​of X / Y / Z triaxial in-situ stress changes, which are obtained synchronously by a fiber Bragg grating stress meter with a range of 0–60 MPa and an accuracy of ±0.5% FS. It is used to quantify the stress state of the rock mass. In a specific example, a fiber Bragg grating stress meter was embedded at the coal-rock interface in Mine A. The sensor range was set to 0–60 MPa and calibrated to ±0.5% FS. The installation location was selected in an area with stable lithology and buried at a depth of 2 m to avoid interference from mining activities. The system synchronized X / Y / Z triaxial data (e.g., maximum principal stress of 25 MPa) in real time and transmitted it via cable. This example demonstrated exceptional embedding accuracy, successfully capturing stress concentration trends during the initial deployment and providing key model input parameters.

[0031] Step 1.4: Collect tunnel deformation and displacement data. Tunnel deformation displacement data: refers to the tunnel convergence deformation, recorded by a MEMS displacement meter with a range of ±300mm and a resolution of 0.01mm, and is used to monitor the stability of the support structure.

[0032] In a specific example, Mine A installed MEMS displacement meters at key nodes of U-shaped steel supports, with a range of ±300mm and a resolution of 0.01mm. Installation was aligned with high-stress areas in the tunnel, and the support interfaces were reinforced. The system recorded deformation in real time (e.g., a daily convergence of 0.5mm), and the data was uploaded via a cable. This example demonstrates the successful identification of early creep through optimized installation locations, providing a displacement benchmark for risk warning.

[0033] Step 1.5: Data access and real-time upload. Explosion-proof armored cable: A protective cable used in underground environments that is explosion-proof and resistant to mechanical damage, ensuring secure sensor data transmission.

[0034] Underground Industrial Ring Network: A high-speed network infrastructure that supports 100Mbps data transmission rate and is used to aggregate sensor data to edge computing nodes in real time.

[0035] In a specific example, Mine A connected acoustic emission, microseismic, stress, and displacement sensors to an underground industrial ring network via explosion-proof armored cables. The system uploaded data at a 100Mbps rate, with edge nodes verifying packet loss in real time (target: <0.1%). After deployment, the initial test achieved a 99.9% data transmission success rate with no interruptions. This implementation focused on network configuration, ensuring minute-level data availability and supporting real-time system alerts.

[0036] Step 2: The collected acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation and displacement data are combined with static geological attribute data to perform time synchronization and spatial coordinate alignment to generate high-dimensional feature cube data with a unified time and space reference; wherein the static geological attribute data includes lithology, structural joints, and strike and dip information.

[0037] Step 2.1: Based on the preset IEEE-1588PTP precise time protocol, the sensor nodes that collect the acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation displacement data are time synchronized to obtain synchronized timestamp data.

[0038] IEEE-1588 PTP protocol: Precision Time Protocol (PTP) achieves sub-millisecond time synchronization through master-slave clock message exchange, solving the problem of clock drift in sensor nodes.

[0039] Synchronous timestamp data: All sensor data carries a unified time tag (format: YYYYMMDD-HHMMSS.FFF), with a time error of ≤0.3ms.

[0040] The underground industrial ring network deploys a PTP master clock, which periodically broadcasts Sync messages to acoustic emission, microseismic, stress, and displacement sensor nodes. Nodes calculate path delays and calibrate their local clocks to ensure that all data stream timestamps are aligned. The master clock generates a reference time tag (e.g., 20250222-022230.123) for subsequent spatiotemporal mapping.

[0041] In a specific example, Mine A deployed a PTP master clock, with sensor nodes connected via Gigabit Ethernet. The master clock broadcasts Sync messages every 2 seconds, and after node calibration, the timestamp error is controlled to ±0.25ms. The acoustic emission waveform data packet is labeled 20250723-092015.456, and the microseismic event data is synchronized to 20250723-092015.458, achieving sub-millisecond alignment across devices.

[0042] Step 2.2: Establish three-dimensional spatial coordinate framework data of the underground engineering area based on the preset RTK-SLAM algorithm and laser scanning point cloud data.

[0043] RTK-SLAM algorithm: It combines real-time kinematic positioning (RTK, accuracy of ±2cm) with simultaneous localization and mapping (SLAM) to construct an absolute 3D coordinate system through lidar point cloud matching.

[0044] Three-dimensional space coordinate frame data: tunnel space model divided into 1m×1m×1m grids, with the coordinate origin being the geodetic reference point.

[0045] The laser SLAM vehicle scans the roadway to generate a high-density point cloud (≥5,000 points / ㎡). The RTK base station provides geodetic coordinate correction. The point cloud is registered to a unified coordinate system, and a 3D grid framework with absolute coordinates is output, providing spatial anchors for monitoring data.

[0046] In a specific example, Mine A used a 32-line LiDAR SLAM vehicle (traveling at 2 km / h) to scan the E1203 working face tunnel, generating a point cloud model. An RTK base station was installed at the tunnel intersection to register the point cloud to the coordinate system origin (X0, Y0, Z0), outputting a grid framework (for example, grid G1205 corresponds to coordinates (120.5, 45.2, -305.3)).

[0047] Step 2.3: performing spatiotemporal mapping on the acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation displacement data based on the synchronous timestamp data and the three-dimensional spatial coordinate frame data to obtain spatially aligned monitoring point data.

[0048] Spatially aligned monitoring point data: the result of mapping the original data to 3D grid nodes, with missing points filled by interpolation.

[0049] Acoustic emission events: The earthquake source is located based on the time difference from the sensor array and mapped to the nearest grid center.

[0050] Microseismic events: The earthquake sources are spherically inverted using the arrival time data of 8 microseismometers and projected onto 3D grid nodes.

[0051] Stress / displacement data: Corresponding grids are associated according to the installation coordinates, and missing points are interpolated using inverse distance weighted (IDW).

[0052] In a specific case, the microseismic event in mining area A was located at (X: 120.7, Y: 45.3, Z: -305.1) through spherical inversion and mapped to grid G1205; the data at the stress gauge installation point (X: 120.5, Y: 45.0, Z: -305.0) were missing and supplemented by IDW interpolation of three adjacent points.

[0053] Step 2.4: Perform sliding window filtering, noise suppression, and energy envelope quantization processing on the acoustic emission waveform data and microseismic waveform data at the edge computing node to obtain preprocessed waveform feature data.

[0054] Sliding window filtering: Segment the waveform with a fixed window (e.g., 10ms) with an overlap rate of 50% to extract local features.

[0055] Energy envelope quantification: Calculate the root mean square (RMS) value of the signal within the window and convert the unit to microjoules (μJ).

[0056] Edge nodes (such as NVIDIA Jetson Orin) perform the following on acoustic emission / microseismic waveforms: Noise reduction: db4 wavelet basis 5-layer decomposition, hard threshold filtering <30dB high-frequency noise; Energy quantification: Calculate the RMS value of each window and output the energy time series curve.

[0057] In a specific example, the edge node divides the acoustic emission waveform into 10ms windows, and the db4 wavelet filters out noise above 32kHz, outputting an RMS energy sequence (e.g., energy value = 120μJ at t = 092015.456).

[0058] Step 2.5: Perform adaptive Kalman filtering to suppress noise and linear interpolation to compensate for missing values ​​on the triaxial stress data and the roadway deformation displacement data to obtain preprocessed stress-displacement characteristic data.

[0059] Adaptive Kalman filtering: Dynamically adjusts noise covariance to suppress temperature drift and transient interference.

[0060] Linear interpolation compensation: Missing points caused by network fluctuations are filled linearly according to adjacent data.

[0061] The stress and displacement series are Kalman filtered to suppress temperature drift, and missing segments (such as those caused by packet loss) are linearly interpolated in time to ensure data continuity.

[0062] In a specific case, the stress data of mining area A was lost for 5 seconds due to network interruption. The system supplemented it by linear interpolation based on the previous value of 25.3MPa and the subsequent value of 25.8MPa. After filtering, the fluctuation range was reduced from ±0.8MPa to ±0.1MPa.

[0063] Step 2.6: Organize the preprocessed waveform feature data, the preprocessed stress-displacement feature data, and the static geological attribute data into high-dimensional feature cube data indexed by time dimension, space dimension, and feature dimension according to a preset time window and spatial resolution.

[0064] High-dimensional feature cube data: a three-dimensional tensor with time (T), space (S), and features (F) as axes (e.g., size T×S×F=100×500×20).

[0065] Time window: 10s as the window and 1s step size sliding package; Spatial resolution: 1m 3 The grid is the smallest unit; Feature dimensions: including waveform energy, stress components, displacement, lithology coding, etc.

[0066] Data is serialized into Apache Arrow format, which supports GPU zero-copy loading.

[0067] In a specific example, mining area A organizes data within a 10-second window into cubes: Time axis: 100 steps (10s ÷ 0.1s); Spatial axis: 500 grids (500m lane); Characteristic axes: acoustic emission energy, triaxial stress, displacement, and sandstone lithology label (code = 3).

[0068] Stored in Arrow format for use by P-Former models.

[0069] Step 3: Input the high-dimensional feature cube data into a preset physical information injection converter for modeling to obtain intermediate feature vector data including energy accumulation rate, stress concentration factor and crack evolution potential, as well as three-dimensional hazard confidence distribution data.

[0070] Step 3.1: Perform frequency domain transformation on the high-dimensional feature cube data to obtain frequency domain feature data.

[0071] Frequency domain feature data: The time domain signal is converted into frequency domain representation through fast Fourier transform (FFT) or short-time Fourier transform (STFT), highlighting the frequency characteristics (such as energy distribution) to facilitate the model to capture periodic or transient events.

[0072] Perform frequency domain transforms on waveform data (acoustic emissions, microseismic data) from high-dimensional feature cubes. FFT is used for global spectral analysis, while STFT is used for sliding window local spectra. The output is a frequency domain tensor (e.g., amplitude spectrum, phase spectrum). This helps identify high-frequency microfractures or low-frequency energy accumulation patterns, providing input for the attention mechanism.

[0073] In a specific case, STFT was applied to acoustic emission waveform data (cube feature axes) at Mine A: the signal was segmented using a 256-point window with 50% overlap, the frequency domain magnitude spectrum was calculated, and a 20-dimensional feature vector was output for frequencies 0–64 kHz. This successfully captured an energy peak in the 15 kHz band, indicating early fracture activity.

[0074] Step 3.2: Input the frequency domain feature data into the multi-head self-attention layer of the physical information injection transformer.

[0075] Multi-head self-attention layer: The core component of Transformer, which calculates the association weights of the input sequence through multiple parallel attention heads (e.g., 8 heads) to enhance feature interaction capabilities.

[0076] Frequency-domain feature data is directly fed into the multi-head self-attention layer of the P-Former. Each attention head independently calculates the similarity of the query, key, and value vectors, preliminarily modeling inter-feature dependencies (such as the temporal correlation between stress and energy). Input dimensions must be aligned with the model (e.g., feature axis normalization) to ensure efficient computation.

[0077] In a specific example, Mine A feeds frequency domain features (size 100×500×128) into an 8-head self-attention layer. Each head processes a different frequency subspace (e.g., head 1 focuses on 0–10 kHz) and outputs preliminary coupling features in preparation for physical injection.

[0078] Step 3.3: When each attention head of the multi-head self-attention layer calculates the similarity between the query vector and the key vector, the rock mass elastic-plastic constitutive residual data and the in-situ stress tensor data are explicitly introduced as penalty terms to calculate the attention weight data that conforms to the mechanical constraints; wherein, the rock mass elastic-plastic constitutive residual data is obtained by calculating the difference between the predicted stress and the actual stress in real time based on the constitutive equation obtained by field measurement or experiment.

[0079] Rock mass elastic-plastic constitutive residual data: the difference between the predicted stress and the actual stress calculated based on the constitutive equation (such as the Hoek-Brown model), reflecting the deviation of mechanical laws.

[0080] In-situ geostress tensor data: three-dimensional stress state measured in situ.

[0081] Penalty term: A constraint term added to the attention energy function to force the model to satisfy energy conservation and volume compatibility.

[0082] In a specific example, the constitutive residual of the sandstone area calculated in mining area A is: the predicted stress is 25.3 MPa. The measured stress is 25.8 MPa, and the residual R ij =0.5MPa; Earth stress tensor S ij Medium σ z =20MPa. Set λ1=0.3 and λ2=0.2 and add them to the attention weight calculation. Result: The model ignores noise interference and focuses on the true energy accumulation event.

[0083] Step 3.4: Perform weighted aggregation on the value vector based on the calculated attention weight data to obtain deep feature coupling data.

[0084] Deep feature coupling data: The output of the weighted summation of the value vector is achieved through attention weights to achieve high-dimensional fusion of cross-temporal and spatial features.

[0085] Based on the attention weights from step 3.3, the Value vectors are weighted and aggregated. Highly weighted features (such as areas of high stress concentration) are enhanced, while low-weight features (such as noise) are suppressed. The output tensor integrates information from multiple sources (waveform, stress, and static properties) to form a unified deep representation.

[0086] In a specific case, the aggregated features of mining area A strengthened the association between microseismic events and stress concentration (weight > 0.8), and the output coupling data was used for expert routing.

[0087] Step 3.5: Input the deep feature coupling data into the multi-expert gating subnetwork of the physical information injection transformer.

[0088] Multi-expert gated sub-network: A structure that includes multiple expert sub-networks (such as hard rock experts and soft rock experts) and a gated network, dynamically adapting to different geological conditions.

[0089] Deep feature coupling data is fed into a gating subnetwork. The gating network selects the most relevant experts based on input attributes (e.g., lithology), resolving cross-scale differences (e.g., brittleness of hard rock vs. plasticity of soft rock). The expert subnetworks are independently trained to handle specific patterns.

[0090] In a specific example, three experts are deployed in Mining Area A: Expert 1 (hard rock), Expert 2 (soft rock), and Expert 3 (jointed rock). The coupled data is first fed into the gated network to prepare for routing.

[0091] Step 3.6: Based on the lithologic attributes of the input data, the multi-expert gating subnetwork dynamically routes the depth feature coupling data to the corresponding expert subnetwork for processing to obtain the intermediate feature vector data and the three-dimensional hazard confidence distribution data.

[0092] Step 3.6.1: Input the depth feature coupling data and the lithology label data in the static geological attribute data into the gating network of the multi-expert gating sub-network.

[0093] Gating network: A neural network that learns the coupling pattern between features and lithology and outputs expert selection weights.

[0094] The depth feature coupling data and lithology labels (derived from static attributes) are fed into a gating network. The network learns the mapping (e.g., sandstone → high energy accumulation pattern) and generates a weight vector.

[0095] In a specific case, mining area A inputs sandstone labels (coding = 3) and coupling features, and the gating network outputs weights [0.7, 0.2, 0.1], indicating that expert 1 is dominant.

[0096] Step 3.6.2: Based on the gated network, learn the coupling patterns between the depth feature coupling data and different lithology types to generate expert selection weight data.

[0097] Expert selection weight data: The probability distribution output by the gating network, indicating to which expert the input data should be routed.

[0098] The gating network generates weights using a softmax function to ensure that the sum is 1. The weights are based on feature-lithology correlations (e.g., high joint density is routed to expert 3).

[0099] In a specific example, the gating network of mining area A learns the “high stress concentration + sandstone” pattern, outputs weights [0.85, 0.10, 0.05], and prefers expert 1.

[0100] Step 3.6.3: Based on the expert selection weight data, dynamically route the deep feature coupling data to the most matching one or more expert sub-networks.

[0101] Dynamic routing: Distribute data to one or more expert subnetworks based on weights.

[0102] Expert selection weights determine the routing path (e.g., experts with weights > 0.5 are selected). Data is distributed in parallel, and expert sub-networks process independently, outputting local features.

[0103] In a specific case, the weights of mining area A are [0.75, 0.20, 0.05], and the data is routed to expert 1 (hard rock expert) to process the fracture evolution potential.

[0104] Step 3.6.4: Process the routed deep feature coupling data based on the selected expert sub-network, and output the intermediate feature vector data and the three-dimensional hazard confidence distribution data.

[0105] Expert sub-network: a specialized neural network (such as CNN or MLP) that processes features under specific geological conditions.

[0106] The selected expert subnetwork processes the input data and outputs intermediate feature vectors (such as energy accumulation rate) and a three-dimensional hazard confidence distribution (grid-level probability map). The confidence threshold is used for subsequent risk assessment.

[0107] In a specific case, Expert 1 of Mine Area A outputs: energy accumulation rate 0.92 (high risk), stress concentration factor 1.5, fracture evolution potential 0.8, and three-dimensional confidence distribution (e.g., confidence of grid G1205 = 0.95).

[0108] Step 4: Based on the intermediate feature vector data, a preset rockburst case knowledge base is retrieved to obtain similar historical case data, and the similar historical case data is spliced ​​and fused with the intermediate feature vector data to generate natural language text data containing explanations of potential inducing mechanisms and recommendations for prevention and control measures.

[0109] Step 4.1: Based on the intermediate feature vector data, a first-stage coarse-grained vector similarity search is performed in the rockburst case knowledge base to obtain a preliminary matching candidate case data set.

[0110] Coarse-grained vector similarity retrieval: Use an approximate nearest neighbor algorithm (such as FAISS) to quickly screen candidate cases and reduce the computational overhead of full-database searches.

[0111] Candidate case dataset: a preliminary matching Top-N historical case set (N ≥ 10), with a similarity threshold > 0.6.

[0112] Intermediate feature vectors (such as energy accumulation rate and stress concentration factor) are input into the FAISSGPU index library, and cosine similarity is calculated with all case vectors in the knowledge base. The top-N candidate cases are returned to ensure coverage of potentially similar scenarios.

[0113] In a specific case, mining area A input the 128-dimensional intermediate feature vector into the FAISS index and retrieved the top-15 candidate cases within 3 milliseconds (the highest similarity was 0.82, corresponding to the E1201 working face rock burst event in 2023).

[0114] Step 4.2: Perform a second-stage fine-grained search on the candidate case data set that has been preliminarily matched. The fine-grained search performs hierarchical matching calculations based on the similarity of static attribute vectors and dynamic sequence vectors to screen out target case data with high similarity. The static attribute vectors include lithology, strike, and joint density, and the dynamic sequence vectors include energy attenuation curves and fracture evolution trajectories.

[0115] Static attribute vector: discrete geological feature encoding of the case (lithology: sandstone = 3, strike: N30°E, joint density: 5 / m 3 ).

[0116] Dynamic sequence vector: the temporal characteristics of the case (energy decay curve: RMS value sequence, crack evolution trajectory: displacement change sequence).

[0117] Hierarchical matching: First calculate the cosine similarity of static attributes, and then evaluate the dynamic sequence consistency through DTW (Dynamic Time Warping) distance.

[0118] Static matching: filter cases with attribute similarity > 0.85; Dynamic matching: Calculate the DTW distance for the remaining cases (the smaller the value, the more similar they are) and select the top-K cases with the smallest distance (K≤5).

[0119] In a specific example, Mine A performs the following on the top 15 candidate cases: Static filtering: cases with joint density differences >10% were eliminated, leaving 8 cases; Dynamic matching: The current crack evolution trajectory has the smallest DTW distance (0.15) with case ID-202405 and is selected as the Top-1 target case.

[0120] Step 4.3: Concatenate the filtered feature representation of the target case data with the intermediate feature vector data to form knowledge-enhanced feature input data.

[0121] Knowledge-enhanced feature input data: The concatenation result of the target case feature and the intermediate feature vector, with the dimension expanded to more than 1.5 times the original feature.

[0122] The static attribute vector (e.g., 128 dimensions) and dynamic sequence vector (e.g., 64 dimensions) of the Top-K target cases are concatenated to the end of the original intermediate feature vector to form a high-dimensional fusion input (e.g., 128+192=320 dimensions).

[0123] In a specific case, Mine A concatenates the top-3 case vector (64 dimensions × 3) with the original vector (128 dimensions) to generate a 320-dimensional feature input and injects historical case context.

[0124] Step 4.4: Input the knowledge-enhanced feature input data into the instruction-fine-tuned language model.

[0125] Instruction fine-tuning language model: LLM fine-tuned based on natural language tasks (such as text generation), mapping feature vectors to text space through adaptation layers.

[0126] The knowledge-enhanced features are converted to the language model embedding dimension (e.g., 768 dimensions) through a linear projection layer, and a prefix instruction (e.g., "Generate rockburst mechanism explanation:") is input to guide the generation direction.

[0127] In a specific example, mining area A used the LLaMA-7B model, added an adaptation layer to map 320-dimensional features to 768 dimensions, and input the command: "Analyze the causes of rockburst based on the following features and provide measures:."

[0128] Step 4.5: Process the knowledge-enhanced feature input data based on the language macro model to automatically generate the natural language text data containing the potential inducing mechanism explanation and prevention and control measures suggestions.

[0129] Natural language text data: structured text containing inducing mechanisms (e.g., "dominated by lateral tectonic stress") and prevention and control measures (e.g., "φ75mm pressure relief hole").

[0130] The language model decodes the generated text and parses the key fields: Inducing mechanism: quantitative attribution (such as stress ratio); Prevention and control measures: specific parameters (aperture, spacing, support specifications); Reference case: Correlating historical event IDs.

[0131] In a specific example, the output text of Mine A is: Triggering mechanism: Transverse tectonic stress (accounting for 68%) superimposed on energy accumulation in brittle sandstone. Prevention and control measures: Reinforced anchor cables (22mm diameter, 1.2m spacing) and pressure relief holes (φ75mm x 6m, 5m x 5m spacing). Reference case: 202405-E0502. Step 5: Generate three-dimensional risk heat map data based on the three-dimensional hazard confidence distribution data, extract potential fracture surface spatial position data based on the three-dimensional risk heat map data, and visualize the three-dimensional risk heat map data, potential fracture surface spatial position data and the natural language text data.

[0132] Step 5.1: Map the three-dimensional hazard confidence distribution data to the spatial voxels corresponding to the three-dimensional roadway geographic information model to generate initial three-dimensional risk heat map data.

[0133] 3D tunnel geographic information model: A 3D digital twin of the tunnel based on the integration of BIM (Building Information Modeling) and GIS (Geographic Information System), which includes spatial information such as support structures and rock formation boundaries.

[0134] Spatial voxel: A 1m×1m×1m cubic grid unit, which serves as the basic unit for risk data mapping.

[0135] Initial 3D risk heat map data: The hazard confidence level (0.0-1.0) is mapped to a gradient color space distribution generated by the corresponding voxel, with red indicating high risk (>0.8) and blue indicating safety (<0.3).

[0136] The 3D hazard confidence distribution data (grid-level probability) output by P-Former is aligned with the spatial voxels of the roadway geographic information model. Each voxel is assigned an RGB color based on the confidence value: Low risk (0.0-0.3): blue; Medium risk (0.3-0.8): yellow; High risk (0.8-1.0): red; Generate an initial heat map for cluster analysis.

[0137] In a specific example, mining area A maps the confidence level 0.95 of grid G1205 to RGB (255, 0, 0), and the confidence level 0.4 of grid G1210 to RGB (255, 255, 0), forming an initial risk heat map.

[0138] Step 5.2: Perform a density and spatial connectivity-based clustering analysis algorithm on the initial three-dimensional risk heat map data to identify high-risk connected area data.

[0139] Density and spatial connectivity clustering: An improvement on the DBSCAN algorithm that combines voxel density thresholds (e.g., 5 adjacent voxels) and spatial connectivity rules (shared faces / edges / vertices) to identify physically continuous high-risk blocks.

[0140] High-risk connected region data: a set of spatial regions output by clustering, each of which contains ≥10 high-risk voxels and satisfies geometric connectivity.

[0141] Perform the following operations on the initial heatmap: Density filtering: only voxels with confidence > 0.8 are retained; Connectivity check: Scan 26 neighborhoods (up, down, left, right, front, back, and diagonal) to mark connected bodies; Area merging: If the distance between two areas is less than 2m, they will be merged into the same high-risk area.

[0142] Outputs the region ID and the list of voxels it contains.

[0143] In a specific case, three high-risk connected areas were identified in Mining Area A: Region ID-R1: contains voxels {G1205, G1206, G1210} (volume 12m 3 ); Region ID-R2: contains voxels {G1301, G1302} (volume 8m 3 ).

[0144] Step 5.3: Calculate the minimum convex hull geometric data for each of the identified high-risk connected area data, where the minimum convex hull geometric data represents the spatial position data of the potential rupture surface.

[0145] Minimum convex hull geometry: The minimum convex polyhedron that encloses all voxels in the high-risk connected area, whose surface represents the spatial location of the potential rupture surface.

[0146] Potential rupture surface spatial location data: a set of three-dimensional coordinates of the convex hull vertices (such as [X1, Y1, Z1], [X2, Y2, Z2]...), defining the rupture surface boundary.

[0147] For each high-risk connected area, perform: Extract the coordinates of the center points of all voxels in the region; Calculate the minimum convex hull using the Quickhull algorithm; Output convex hull vertex coordinate sequence and triangle patch topology.

[0148] The convex hull surface is the potential energy release fracture surface.

[0149] In a specific example, mining area A calculates the convex hull for region ID-R1: Vertex coordinates: [(120.5, 45.2, -305.3), (120.8, 45.5, -305.1), (121.0, 45.0, -305.5)]; Fracture surface area: 15.2m 2 .

[0150] Step 5.4: Render and display the three-dimensional tunnel scene superimposed with the initial three-dimensional risk heat map data, the convex hull boundary of the high-risk connected area data, and the natural language text data on a three-dimensional visualization platform.

[0151] Natural language text data: explanation of the inducing mechanism and prevention and control measures generated in step 4 (e.g., "lateral structural stress dominates, φ75mm pressure relief hole is recommended").

[0152] Implement four-layer overlay rendering on the Three.js or CesiumJS platform: Basic scene: roadway BIM-GIS model (grey and semi-transparent); Risk heat: The initial heat map is rendered with a semi-transparent volume (red → blue gradient); Fracture surface boundary: line connecting convex hull vertices (thick green solid line); Text annotation: Floating knowledge cards associated with high-risk areas.

[0153] Users can interactively explore details by cutting and rotating.

[0154] In a specific example, the A Mine Dispatching screen displays: Region ID-R1: red semi-transparent thermal volume + green convex hull boundary; Click on the pop-up text: "Induced mechanism: transverse stress (68%) + brittle sandstone; measures: φ75mm pressure relief holes with a spacing of 5m"; Supports Z-axis sectioning to observe deep stress distribution.

[0155] Step 6: Perform risk assessment based on the three-dimensional hazard confidence distribution data, trigger an edge alarm signal when the risk level reaches a preset threshold, and record on-site execution feedback data for the prevention and control measures recommended.

[0156] Step 6.1: Perform real-time assessment on the risk level of the three-dimensional hazard confidence distribution data within a continuous time window.

[0157] Risk level: Warning levels divided by confidence level (Level I: 0.6-0.7, Level II: 0.7-0.8, Level III: >0.8), each level corresponds to a different response strategy.

[0158] Time window: The system defaults to 10 seconds as the evaluation unit, and continuously monitors the confidence change trend.

[0159] For each spatial grid's confidence data, perform: Instantaneous rating: matches the risk level based on the current value (e.g. 0.85 → Level III); Trend analysis: Calculate the confidence slope of three consecutive windows (e.g. >0.02 / window indicates accelerated risk).

[0160] In a specific case, the confidence level of grid G1205 in mining area A for three consecutive windows is: 0.82→0.85→0.88, with a slope of 0.03 / window, and it is determined to be a level III acceleration risk.

[0161] Step 6.2: When it is detected that the risk levels of a predetermined number of consecutive time windows all reach or exceed the preset risk threshold level, it is determined that the alarm triggering condition is met.

[0162] Continuous preset number: The system preset continuous alarm threshold (e.g. Level III requires 2 consecutive windows ≥ 0.8).

[0163] Dynamic detection of two conditions: Current level: target grid confidence ≥ threshold (e.g., level III > 0.8); Duration: N consecutive windows meet the condition (N is configurable).

[0164] If both conditions are met, an alarm is triggered.

[0165] In a specific example, a Level III alarm in Mine A requires two consecutive windows with a confidence level of ≥ 0.8. The trigger condition is met when the confidence level of grid G1301 in two consecutive windows exceeds 0.83 and then 0.86.

[0166] Step 6.3: When the alarm triggering condition is met, the edge alarm signal is immediately sent to the edge alarm device deployed at the tunnel site to trigger an audible and visual alarm.

[0167] Edge alarm device: A hardware device (buzzer + three-color LED) deployed on-site in the tunnel, supporting audible and visual alarms (flashing red light + 105dB buzzer).

[0168] Alarm signals are sent via the underground industrial ring network in milliseconds: Device positioning: Signals are associated with high-risk grid coordinates (e.g., G1205); Graded response: Level III triggers full-power sound and light, Level II only flashes the yellow light.

[0169] In a specific case, G1205 in mining area A triggered a Level III alarm, and the red light of the nearest edge device (coordinates 120.5, 45.2, -305.3) flashed at a high frequency, and the buzzer sounded every 3 seconds.

[0170] Step 6.4: An alarm information window automatically pops up on the central dispatch platform, and the alarm information window includes a high-risk area identifier, the prevention and control measures recommendations, and key explanations in the natural language text data.

[0171] Alarm information window: An interactive interface that pops up automatically on the dispatching platform, including three-dimensional positioning, prevention and control recommendations, and mechanism explanations.

[0172] The window content is presented in a structured manner, for example: [High-risk area] E12-3# (G1205-G1210);

Confidence

Prevention and control measures

Mechanism

[0173] Step 6.5: Collect and record the type of operation, operation time, and post-operation regional stability observation data performed by on-site personnel according to the prevention and control measures recommended to form the execution feedback data.

[0174] Execution feedback data: structured record fields (operation type, timestamp, stability indicators).

[0175] On-site personnel recorded through the explosion-proof mobile APP: Operation type: predefined options (anchor reinforcement / pressure relief drilling / shotcrete support); Effect observation: Select the stability change from the drop-down menu (significant improvement / partial improvement / no change).

[0176] Data is automatically synchronized to the central platform.

[0177] In a specific example, Mine A recorded: "20250723-0920|6 pressure relief holes @5m spacing|Energy dropped by 60% after 2h|Significant improvement".

[0178] Step 7: Monitor the input distribution characteristics of the high-dimensional feature cube data and the physical residual distribution characteristics generated by the modeling process, determine the model drift state based on the monitoring results, and trigger the incremental fine-tuning process of the model according to the model drift state and the execution feedback data.

[0179] Step 7.1: Calculate in real time first distribution difference measurement data between the input distribution features of the current high-dimensional feature cube data and the historical benchmark input distribution features.

[0180] Input distribution characteristics of high-dimensional feature cube data: refers to the statistical properties of the feature cube data generated in step 2.6 (such as mean and variance), which are used to describe the distribution status of real-time monitoring data.

[0181] Historical baseline input distribution characteristics: The data distribution baseline values ​​recorded by the system during stable operation (such as after initial training) serve as a comparison reference.

[0182] First distribution difference metric data: an indicator that quantifies the difference between the current input distribution and the historical benchmark, usually calculated using the Kullback-Leibler (KL) divergence. A larger value indicates a more significant difference.

[0183] Extract the distribution characteristics (such as the mean and standard deviation of each feature channel) of high-dimensional feature cube data (such as a tensor of time × space × feature dimensions) in real time. Compare with the historical benchmark (distribution parameters stored in the database) and calculate the KL divergence: The KL divergence formula simplifies to a difference value (0 means no difference, >0.2 means significant drift).

[0184] The system updates the difference value every 10 minutes to ensure real-time performance.

[0185] This helps identify changes in the input data distribution (such as new rock formations or sensor drift) and prevents models from becoming invalid due to data drift.

[0186] In a specific example, Mining Area A calculated the KL divergence between the current input distribution (feature cube mean = 25.3, standard deviation = 5.1) and the historical benchmark (mean = 24.8, standard deviation = 4.9). The difference was 0.18, below the threshold of 0.2, and no alarm was triggered. This example focuses on the KL divergence calculation process.

[0187] Step 7.2: Calculate in real time the second distribution difference measurement data between the physical residual distribution characteristics generated by the current modeling process and the expected physical residual distribution characteristics.

[0188] The physical residual distribution characteristics generated by the modeling process refer to the statistical properties of the rock mass elastic-plastic constitutive residuals in the output of the P-Former model (step 3) (such as the distribution form of the residual values), which reflect whether the model conforms to the laws of mechanics.

[0189] Expected physical residual distribution characteristics: The ideal residual distribution defined during the model training phase (such as a Gaussian distribution with a mean close to 0) indicates that the model strictly satisfies physical constraints.

[0190] Second distribution difference metric data: an indicator that quantifies the difference between the current residual distribution and the expected distribution, also calculated using KL divergence or Jensen-Shannon distance.

[0191] Monitor the physical residuals generated during the P-Former inference process (e.g., the difference between the constitutive equation prediction and the actual stress). Extract the current residual distribution characteristics (e.g., the histogram statistics of the residual sequence) and compare them with the expected distribution (parameters saved during training): Calculate the difference index to ensure that the model reasoning process always follows the laws of rock mechanics.

[0192] A high difference value indicates that the physical constraints of the model are invalid (e.g., inaccurate prediction of crack evolution).

[0193] Combined with step 7.1, this provides a dual drift detection mechanism.

[0194] In a specific example, the Jensen-Shannon distance is calculated between the current residual distribution (mean = 0.5 MPa, variance = 0.2) and the expected distribution (mean = 0.1 MPa, variance = 0.1) for Mining Area A. The difference is 0.25, exceeding the threshold of 0.2. This example focuses on a single operating point for residual distribution comparison.

[0195] Step 7.3: Calculate comprehensive drift index data based on the first distribution difference measurement data and the second distribution difference measurement data.

[0196] Comprehensive drift index data: A weighted composite value (e.g., weighted average) combining the first and second distribution difference metrics is used to comprehensively assess the model drift status. The formula is: Comprehensive index = w1 × first difference + w2 × second difference. The weights w1 and w2 are set based on experience (e.g., w1 = 0.6, w2 = 0.4).

[0197] The system automatically performs a weighted summation of the first distribution difference measure (input data drift) and the second distribution difference measure (physical residual drift): The weights are calibrated based on historical data (e.g., if input data drift has a greater impact on the model, the weights are higher).

[0198] Outputs a composite drift index (range 0-1), where values ​​> 0.3 indicate a high risk of drift.

[0199] This provides a global view of drift and avoids misjudgment of a single indicator.

[0200] In a specific example, Mining District A sets weights w1 = 0.6 and w2 = 0.4. The first difference is 0.18, and the second difference is 0.25. The calculated comprehensive index is 0.6 × 0.18 + 0.4 × 0.25 = 0.208, which is lower than the threshold of 0.3.

[0201] Step 7.4: When the comprehensive drift index data continuously exceeds the preset drift threshold for a predetermined period of time, it is determined that model drift has occurred, triggering the model downgrade operation, automatically switching to the historical stable model version for inference, and generating model aging alarm information.

[0202] Preset drift threshold: The system-defined upper limit of drift tolerance (such as comprehensive index > 0.3), determined based on historical testing.

[0203] Predetermined duration: The time period for which the threshold is exceeded continuously (e.g., 1 hour) to avoid false triggering due to instantaneous fluctuations.

[0204] Model downgrade operation: Automatically switch to the historical stable model version (such as last month's backup) to ensure that the system continues to run.

[0205] Model aging alarm information: a structured alarm message (such as the alarm code "MODEL-DRIFT-ALERT") sent to the operation and maintenance platform.

[0206] Monitor comprehensive drift indicators: If the value exceeds a threshold (e.g., 0.3) for one consecutive hour, the model is considered to be drifting.

[0207] The system immediately switches to a historically stable model (such as a backup in ONNX format) and sends an alarm.

[0208] During the downgrade period, the backup model is used for inference to avoid invalid risk warnings.

[0209] This ensures high system availability and complies with the "drift detector" design in the technical solution.

[0210] In a specific case, the comprehensive index of mining area A was > 0.35 for one consecutive hour, triggering a downgrade: switching to the historical model version V2.1, and sending the alarm code "DRIFT-ALERT-001" to the operation and maintenance end.

[0211] Step 7.5: Collect the execution feedback data, model prediction error data, and actual alarm event data, and annotate them to form an incremental training sample data set.

[0212] Execution feedback data: Operation type, timestamp, and stability results (e.g., “Energy dropped by 60% after pressure relief hole construction”) recorded in step 6.5.

[0213] Model prediction error data: Quantified value of the difference between model output and actual events (such as rock burst occurrence).

[0214] Actual alarm event data: records the details of the actual event that triggered the alarm.

[0215] Incremental training sample dataset: a labeled dataset containing input features and target labels (such as corrected hazard confidence).

[0216] Collect and integrate three data sources: Execution feedback: The effectiveness of measures input by on-site personnel through the APP.

[0217] Prediction error: Comparing model output with actual monitoring data.

[0218] Alarm Events: Stores alarm trigger logs.

[0219] The labeled dataset is used for incremental training to ensure that samples cover new scenarios.

[0220] In a specific example, data from Mine A was collected: the execution feedback was "stress reduction of 30% after anchor cable reinforcement," the prediction error was 0.1, and the alarm event ID was -2025. After annotation, a sample set was formed, containing 100 new entries. This example focuses on the data collection and annotation process.

[0221] Step 7.6: Based on the incremental training sample data set, when a degradation of model performance is detected or a regular update instruction is received, an incremental fine-tuning training process of the physical information injection transformer is triggered to update the model parameters.

[0222] Incremental fine-tuning training process: Based on the existing model weights, a new dataset is used for a limited number of training rounds (such as 10 rounds) to update the parameters without retraining the entire model.

[0223] Model performance degradation: evaluated by validation set accuracy or residual metrics (e.g., accuracy degradation > 5%).

[0224] Regular update instructions: a system-preset timed trigger mechanism (such as once a week).

[0225] When performance degradation is detected or a command is received: Load the incremental dataset and fine-tune P-Former on the GPU server.

[0226] Optimize multi-objective loss functions (such as reconstruction error + physical residual penalty).

[0227] The updated model is deployed back to the edge node to support federated learning sharing.

[0228] In a specific example, the detection and verification accuracy of mining area A dropped by 7%, triggering fine-tuning: P-Former was trained using an incremental dataset (200 samples). After 10 rounds, the model was updated and federated learning was synchronized to other mining areas. This example focuses on performance testing and fine-tuning triggers.

[0229] The above is an embodiment of the method proposed in this application. Based on the same inventive concept, this application embodiment also provides a rockburst early identification device based on physical information injection, the structure of which is as follows: Figure 2 shown.

[0230] Figure 2 This is a schematic diagram of the internal structure of a rockburst early identification device based on physical information injection provided in an embodiment of the present application. Figure 2 As shown, the equipment includes: at least one processor 201; and, a memory 202 communicatively coupled to the at least one processor; The memory 202 stores instructions that can be executed by at least one processor, and the instructions are executed by the at least one processor 201 to enable the at least one processor 201 to: Acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data are collected in the underground engineering area; the collected acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data are combined with static geological attribute data for time synchronization and spatial coordinate alignment to generate high-dimensional feature cube data with a unified time and space reference; wherein, the static geological attribute data includes lithology, structural joints and strike and dip information; the high-dimensional feature cube data is input into a preset physical information injection converter for modeling to obtain intermediate feature vector data including energy accumulation rate, stress concentration factor and crack evolution potential and three-dimensional hazard confidence distribution data; based on the intermediate feature vector data, a preset rock burst case knowledge base is retrieved to obtain similar historical case data, and the similar historical case data is compared with the intermediate feature vector data. The three-dimensional hazard confidence distribution data are spliced ​​and fused to generate natural language text data containing explanations of potential inducing mechanisms and suggestions for prevention and control measures; three-dimensional risk heat map data are generated based on the three-dimensional hazard confidence distribution data, potential fracture surface spatial position data are extracted based on the three-dimensional risk heat map data, and the three-dimensional risk heat map data, potential fracture surface spatial position data and the natural language text data are visualized; risk assessment is performed based on the three-dimensional hazard confidence distribution data, an edge alarm signal is triggered when the risk level reaches a preset threshold, and execution feedback data on site for the prevention and control measures is recorded; the input distribution characteristics of the high-dimensional feature cube data and the physical residual distribution characteristics generated by the modeling processing are monitored, the model drift state is judged based on the monitoring results, and the incremental fine-tuning process of the model is triggered according to the model drift state and the execution feedback data.

[0231] Some embodiments of the present application provide corresponding Figure 1 A non-volatile computer storage medium for early identification of rockburst based on physical information injection is provided, wherein computer executable instructions are stored, and the computer executable instructions are set as follows: Acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data are collected in the underground engineering area; the collected acoustic emission waveform data, microseismic waveform data, triaxial stress data and tunnel deformation and displacement data are combined with static geological attribute data for time synchronization and spatial coordinate alignment to generate high-dimensional feature cube data with a unified time and space reference; wherein, the static geological attribute data includes lithology, structural joints and strike and dip information; the high-dimensional feature cube data is input into a preset physical information injection converter for modeling to obtain intermediate feature vector data including energy accumulation rate, stress concentration factor and crack evolution potential and three-dimensional hazard confidence distribution data; based on the intermediate feature vector data, a preset rock burst case knowledge base is retrieved to obtain similar historical case data, and the similar historical case data is compared with the intermediate feature vector data. The three-dimensional hazard confidence distribution data are spliced ​​and fused to generate natural language text data containing explanations of potential inducing mechanisms and suggestions for prevention and control measures; three-dimensional risk heat map data are generated based on the three-dimensional hazard confidence distribution data, potential fracture surface spatial position data are extracted based on the three-dimensional risk heat map data, and the three-dimensional risk heat map data, potential fracture surface spatial position data and the natural language text data are visualized; risk assessment is performed based on the three-dimensional hazard confidence distribution data, an edge alarm signal is triggered when the risk level reaches a preset threshold, and execution feedback data on site for the prevention and control measures is recorded; the input distribution characteristics of the high-dimensional feature cube data and the physical residual distribution characteristics generated by the modeling processing are monitored, the model drift state is judged based on the monitoring results, and the incremental fine-tuning process of the model is triggered according to the model drift state and the execution feedback data.

[0232] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from the other embodiments. In particular, the IoT device and media embodiments are generally similar to the method embodiments, so their description is relatively simple. For relevant portions, refer to the description of the method embodiments.

[0233] The system and medium provided in the embodiments of the present application correspond one-to-one to the method. Therefore, the system and medium also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system and medium will not be repeated here.

[0234] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0235] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0236] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0237] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0238] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0239] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0240] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0241] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0242] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for early identification of rockburst based on physical information injection, characterized in that: The method comprises: Collect acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation and displacement data within the underground engineering area; The collected acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation and displacement data are combined with static geological attribute data to perform time synchronization and spatial coordinate alignment to generate high-dimensional feature cube data with a unified time and space reference; wherein the static geological attribute data includes lithology, structural joints, and strike and dip information; Inputting the high-dimensional characteristic cube data into a preset physical information injection converter for modeling, obtaining intermediate characteristic vector data including energy accumulation rate, stress concentration factor and crack evolution potential, as well as three-dimensional hazard confidence distribution data; Based on the intermediate feature vector data, a preset rockburst case knowledge base is searched to obtain similar historical case data, and the similar historical case data is spliced ​​and fused with the intermediate feature vector data to generate natural language text data containing explanations of potential inducing mechanisms and recommendations for prevention and control measures; generating three-dimensional risk heat map data based on the three-dimensional hazard confidence distribution data, extracting potential fracture surface spatial position data based on the three-dimensional risk heat map data, and visually presenting the three-dimensional risk heat map data, the potential fracture surface spatial position data, and the natural language text data; Perform risk assessment based on the three-dimensional hazard confidence distribution data, trigger an edge warning signal when the risk level reaches a preset threshold, and record on-site execution feedback data for the prevention and control measures recommended; Monitor the input distribution characteristics of the high-dimensional feature cube data and the physical residual distribution characteristics generated by the modeling process, determine the model drift state based on the monitoring results, and trigger the incremental fine-tuning process of the model according to the model drift state and the execution feedback data.

2. The method for early identification of rockburst based on physical information injection according to claim 1, characterized in that: The collected acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation and displacement data are combined with static geological attribute data to perform time synchronization and spatial coordinate alignment to generate high-dimensional feature cube data with a unified time and space reference, specifically including: Based on the preset IEEE-1588PTP precise time protocol, the sensor nodes that collect the acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation and displacement data are time-synchronized to obtain synchronized timestamp data; Establishing three-dimensional spatial coordinate framework data of the underground engineering area based on a preset RTK-SLAM algorithm and laser scanning point cloud data; Performing spatiotemporal mapping on the acoustic emission waveform data, microseismic waveform data, triaxial stress data, and tunnel deformation and displacement data based on the synchronized timestamp data and the three-dimensional spatial coordinate frame data to obtain spatially aligned monitoring point data; Performing sliding window filtering, noise suppression, and energy envelope quantization processing on the acoustic emission waveform data and the microseismic waveform data at the edge computing node to obtain preprocessed waveform feature data; Performing adaptive Kalman filtering to suppress noise and linear interpolation to compensate for missing values ​​on the triaxial stress data and the roadway deformation displacement data to obtain preprocessed stress-displacement characteristic data; The preprocessed waveform feature data, the preprocessed stress-displacement feature data and the static geological attribute data are organized into high-dimensional feature cube data indexed by time dimension, space dimension and feature dimension according to a preset time window and spatial resolution.

3. The method for early identification of rockburst based on physical information injection according to claim 1, characterized in that: The high-dimensional feature cube data is input into a preset physical information injection converter for modeling to obtain intermediate feature vector data including energy accumulation rate, stress concentration factor and crack evolution potential, as well as three-dimensional hazard confidence distribution data, specifically including: Performing frequency domain transformation processing on the high-dimensional feature cube data to obtain frequency domain feature data; Inputting the frequency domain feature data into the multi-head self-attention layer of the physical information injection transformer; When each attention head of the multi-head self-attention layer calculates the similarity between the query vector and the key vector, the rock mass elastic-plastic constitutive residual data and the in-situ ground stress tensor data are explicitly introduced as penalty terms to calculate the attention weight data that meets the mechanical constraints; wherein the rock mass elastic-plastic constitutive residual data is obtained by calculating the difference between the predicted stress and the actual stress in real time based on the constitutive equation obtained by field measurement or experiment; Performing weighted aggregation on the value vector based on the calculated attention weight data to obtain deep feature coupling data; Inputting the deep feature coupling data into the multi-expert gating subnetwork of the physical information injection transformer; Based on the lithologic attributes of the input data, the multi-expert gating subnetwork dynamically routes the depth feature coupling data to the corresponding expert subnetwork for processing to obtain the intermediate feature vector data and the three-dimensional hazard confidence distribution data.

4. The method for early identification of rockburst based on physical information injection according to claim 3, characterized in that: Based on the lithologic attributes of the input data, the multi-expert gating subnetwork dynamically routes the depth feature coupling data to the corresponding expert subnetwork for processing to obtain the intermediate feature vector data and the three-dimensional hazard confidence distribution data, specifically including: Inputting the depth feature coupling data and the lithology label data in the static geological attribute data into the gating network of the multi-expert gating sub-network; Learning the coupling patterns between the depth feature coupling data and different lithologic types based on the gating network to generate expert selection weight data; Dynamically routing the deep feature coupling data to the most matching one or more expert sub-networks based on the expert selection weight data; The selected expert sub-network processes the routed deep feature coupling data and outputs the intermediate feature vector data and the three-dimensional hazard confidence distribution data.

5. The method for early identification of rockburst based on physical information injection according to claim 1, characterized in that: Based on the intermediate feature vector data, a preset rockburst case knowledge base is searched to obtain similar historical case data, and the similar historical case data is spliced ​​and fused with the intermediate feature vector data to generate natural language text data containing explanations of potential inducing mechanisms and recommendations for prevention and control measures, specifically including: Based on the intermediate feature vector data, a first-stage coarse-grained vector similarity search is performed in the rockburst case knowledge base to obtain a preliminary matching candidate case data set; A second stage of fine-grained retrieval is performed on the candidate case data set that has been initially matched. The fine-grained retrieval performs hierarchical matching calculations based on the similarity of static attribute vectors and dynamic sequence vectors to screen out target case data with high similarity. The static attribute vectors include lithology, strike, and joint density, and the dynamic sequence vectors include energy decay curves and fracture evolution trajectories. splicing the filtered feature representation of the target case data with the intermediate feature vector data to form knowledge-enhanced feature input data; Inputting the knowledge-enhanced feature input data into the instruction-fine-tuned language model; The knowledge-enhanced feature input data is processed based on the language macro model to automatically generate the natural language text data containing the potential inducing mechanism explanation and prevention and control measures suggestions.

6. The method for early identification of rockburst based on physical information injection according to claim 1, characterized in that: Generating three-dimensional risk heat map data based on the three-dimensional hazard confidence distribution data, extracting potential fracture surface spatial position data based on the three-dimensional risk heat map data, and visually presenting the three-dimensional risk heat map data, the potential fracture surface spatial position data, and the natural language text data, specifically including: Mapping the three-dimensional hazard confidence distribution data to spatial voxels corresponding to the three-dimensional roadway geographic information model to generate initial three-dimensional risk heat map data; Performing a cluster analysis algorithm based on density and spatial connectivity on the initial three-dimensional risk heat map data to identify high-risk connected area data; Calculating minimum convex hull geometric data for each of the identified high-risk connected area data, wherein the minimum convex hull geometric data represents the spatial position data of the potential rupture surface; A three-dimensional tunnel scene is rendered and displayed on a three-dimensional visualization platform, which superimposes the initial three-dimensional risk heat map data, the convex hull boundary of the high-risk connected area data, and the natural language text data.

7. The method for early identification of rockburst based on physical information injection according to claim 1, characterized in that: A risk assessment is performed based on the three-dimensional hazard confidence distribution data. When the risk level reaches a preset threshold, an edge warning signal is triggered, and on-site feedback data on the implementation of the prevention and control measures is recorded, including: Performing real-time assessment of the risk level of the three-dimensional hazard confidence distribution data within a continuous time window; When it is detected that the risk level of a predetermined number of consecutive time windows reaches or exceeds the preset risk threshold level, it is determined that the alarm triggering condition is met; When the alarm triggering condition is met, the edge alarm signal is immediately sent to the edge alarm device deployed at the tunnel site to trigger an audible and visual alarm; Automatically pop up an alarm information window on the central dispatch platform, the alarm information window including the high-risk area identification, the prevention and control measures recommendations and key explanations in the natural language text data; Collect and record the operation type, operation time and post-operation regional stability observation results data performed by on-site personnel according to the prevention and control measures recommended to form the execution feedback data.

8. The method for early identification of rockburst based on physical information injection according to claim 1, characterized in that: Monitoring the input distribution characteristics of the high-dimensional feature cube data and the physical residual distribution characteristics generated by the modeling process, judging the model drift state based on the monitoring results, and triggering the incremental fine-tuning process of the model according to the model drift state and the execution feedback data, specifically including: Calculating in real time first distribution difference measurement data between the input distribution features of the current high-dimensional feature cube data and the historical benchmark input distribution features; Calculating in real time second distribution difference measurement data between the physical residual distribution characteristics generated by the current modeling process and the expected physical residual distribution characteristics; Calculating comprehensive drift index data based on the first distribution difference metric data and the second distribution difference metric data; When the comprehensive drift index data exceeds the preset drift threshold continuously for a predetermined period of time, it is determined that model drift has occurred, triggering the model downgrade operation, automatically switching to the historical stable model version for inference, and generating model aging alarm information; Collecting the execution feedback data, model prediction error data, and actual alarm event data, and annotating them to form an incremental training sample data set; Based on the incremental training sample data set, when a degradation of model performance is detected or a regular update instruction is received, an incremental fine-tuning training process of the physical information injection converter is triggered to update model parameters.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and operable on the processor, wherein: When the processor executes the computer program, the method for early identification of rockburst based on physical information injection according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program implements a method for early identification of rockburst based on physical information injection according to any one of claims 1 to 8.

Images