A mine time-delay rock burst dynamic short-term intelligent prediction system

Abstract

The application discloses a mine time-delay rock burst dynamic short-term intelligent prediction system, and belongs to the field of rock burst disaster short-term prediction. The system comprises a deep well three-dimensional microseismic field global perception module, a microseismic characteristic parameter extraction module, a rock burst disaster evolution data set construction module, a data preprocessing module, a deep learning prediction module, an explainability analysis module and a dynamic short-term intelligent prediction module. A dynamic tracking type microseismic sensing network is constructed through a three-stage annular nested layout strategy, microseismic event data is collected in real time and space-time characteristic parameters are extracted, a rock burst time series data set is constructed in combination with a sliding time window method, and the occurrence probability and intensity grade of rock burst are predicted. The system adopts a grey wolf-simulated annealing hybrid algorithm to optimize model parameters, and the credibility of the prediction result is improved through SHAP value analysis. The system can realize dynamic short-term early warning of rock burst disasters and provide scientific decision support for mine safety management.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of rock burst disaster short-impending prediction, and more particularly to a mine time-lag type rock burst dynamic short-impending intelligent prediction system. BACKGROUND

[0002] Rock burst disaster in deep metal mines is one of the common major safety hazards in underground engineering, which has a serious threat to personnel life safety and engineering facilities due to its suddenness and destructiveness. Rock burst disaster short-impending prediction is still a problem in the international mining field. Especially in deep well environment, the complex geological conditions and high stress environment make the research work of rock burst prediction more challenging and urgent.

[0003] Rock burst disaster is often accompanied by microseismic events, and through microseismic monitoring technology, a large amount of microseismic time series data containing microseismic event source location, released energy, moment magnitude and stress drop can be obtained. There is a lot of precursor information of rock burst disaster in the time evolution of microseismic data, and deep mining of its spatio-temporal evolution characteristics is helpful to the prediction of rock burst disaster. When real-time microseismic monitoring data is obtained, it is difficult to quickly obtain the disaster evolution law from the massive data through artificial analysis, and to make accurate rock burst prediction. Therefore, through deep learning algorithm, the disaster characteristics in real-time microseismic data are automatically extracted, and the rock burst disaster is quickly and accurately predicted.

[0004] In the prior art, the intelligent prediction method of metal mine rock burst based on microseismic data can only conduct post-disaster statistical analysis on the microseismic data obtained after the rock burst event, ignoring the microseismic data evolution process in the rock burst disaster occurrence process, which causes the defects that it is difficult to give dynamic prediction and early warning of the disaster in the whole process before the rock burst disaster occurs in the real-time monitoring of deep well engineering site. Therefore, how to reasonably construct a microseismic real-time monitoring data set with rock burst intensity classification or occurrence probability and other warning indicating prediction labels in deep well, make full use of microseismic real-time data to build and train an intelligent model that can give dynamic prediction and accurate prediction of rock burst disaster in real-time monitoring scene, conduct feature analysis by combining statistical machine learning method after obtaining the model and good prediction evaluation, explain the black box structure of deep learning model, and finally build a more efficient and transparent deep metal mine rock burst real-time dynamic short-impending intelligent prediction method with stronger generalization ability. In order to adapt to the urgent needs of real-time monitoring, dynamic prediction and short-impending prediction of deep metal mine rock burst, it becomes a technical problem to be solved at present. SUMMARY

[0005] The application provides a time-delay type rock burst dynamic short-term intelligent prediction system, device and equipment based on microseismic real-time monitoring data, and relates to the technical field of mine rock burst monitoring and early warning.

[0006] In order to achieve the above-mentioned purpose, the application is realized by adopting the following technical scheme:

[0007] The system comprises the following modules:

[0008] The deep well three-dimensional microseismic field global perception module adopts a three-level ring nested layout strategy to construct a dynamic tracking type microseismic sensing network topology structure based on the distribution characteristics of mine roadways and the mechanical parameters of deep rock mass, wherein high-density three-component intelligent microseismic sensors are arranged in the key monitoring area, adaptive azimuth calibration arrays are configured in the transition area, and wideband microseismic event capture nodes are arranged in the peripheral area, and each node is integrated with an edge computing module for collecting and filtering the original microseismic waveform signals;

[0009] The microseismic characteristic parameter extraction module extracts time dimension characteristic parameters and space dimension characteristic parameters, including the number of microseismic events, microseismic event energy, apparent stress, stress drop, microseismic event density and source location, by collecting microseismic signal waveform data in real time;

[0010] The rock burst disaster evolution data set construction module constructs a microseismic space-time characteristic parameter time series data set with rock burst intensity classification or occurrence probability labels based on a sliding time window method;

[0011] The data preprocessing module adopts an adaptive logarithmic function to standardize the data and optimize the time series data feature space;

[0012] The deep learning prediction module uses a deep learning model with a 3D-TCN and BiGRU coupled Attention mechanism, combines a grey wolf-simulated annealing hybrid algorithm to optimize the model structure parameters, and predicts the rock burst occurrence probability and intensity level within 12 hours in the future;

[0013] The explainability analysis module improves the credibility of the system prediction logic by transparently analyzing the relationship between the model input and output;

[0014] The dynamic short-term intelligent prediction module realizes real-time dynamic prediction and early warning of rock burst disasters based on the above modules.

[0015] In one scheme, the three-level ring-shaped nested layout strategy in the deep well three-dimensional microseismic field global perception module includes a key monitoring area, a transition area and a peripheral area, wherein the sensor sampling rate of the key monitoring area is greater than or equal to 10 kHz, the sensor array of the transition area contains a 6-degree-of-freedom attitude sensor, and the sampling frequency range of the wideband microseismic event capture node of the peripheral area is 1-1 kHz.

[0016] In one scheme, the time dimension characteristic parameters extracted in the microseismic characteristic parameter extraction module include the number of microseismic events, microseismic event energy, apparent stress, apparent volume, volume change potential, stress drop and moment magnitude mutation index, and the spatial dimension characteristic parameters include microseismic event density, event density space-time fractal dimension and focal position.

[0017] In one scheme, the rockburst catastrophic evolution data set construction module constructs a time series data set by a sliding time window method, inputs 720 time stamp data in the past 12 hours, and outputs rockburst occurrence probability labels and intensity level labels in the future 12 hours, wherein the intensity level is divided into no rockburst, slight rockburst, moderate rockburst and severe rockburst.

[0018] In one scheme, the data preprocessing module adopts an adaptive logarithmic function to standardize the data, adjusts the scaling constant and parameter to avoid distortion effect caused by too large numerical difference, and optimizes the data distribution to adapt to the input of the deep learning model.

[0019] In one scheme, the deep learning prediction module adopts 3D-TCN to extract the space-time local features of microseismic parameters, and captures long sequence features in space-time dimensions through BiGRU combined with an Attention mechanism, and finally outputs the rockburst occurrence probability and intensity level.

[0020] In one scheme, the deep learning prediction module optimizes the model structure parameters through a grey wolf-simulated annealing hybrid algorithm, wherein the grey wolf algorithm is used for global search of structure parameters, the simulated annealing mechanism is used for local fine optimization, and the target function is based on prediction accuracy and model generalization ability.

[0021] In one scheme, the explainability analysis module adopts a SHAP value analysis method to reveal the mapping relationship between the microseismic characteristic parameters and the rockburst prediction results, and improves the transparency and credibility of the deep learning model.

[0022] In one scheme, the dynamic short-term intelligent prediction module combines real-time microseismic field monitoring data and deep learning prediction results to provide dynamic early warning information of rockburst disasters, including rockburst occurrence probability and intensity level, and provides decision support for mine safety management.

[0023] The present application has the following beneficial effects:

[0024] By using deep learning models to perform online analysis of real-time microseismic monitoring time series data, dynamic short-term forecasting of rockburst disasters can be achieved.

[0025] The model can automatically and efficiently capture subtle changes in the evolution of microseismic data, improving the accuracy and reliability of rockburst disaster prediction.

[0026] By combining statistical machine learning methods to deconstruct and analyze deep learning models, the prediction models become more transparent, reliable, and interpretable.

[0027] The systematic approach to acquiring real-time microseismic monitoring data, constructing dynamic datasets for rockburst disasters, and verifying, optimizing, and implementing dynamic short-term rockburst prediction models is applicable to deep metal mines under different geological conditions and has good generalization ability. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the system construction process of the present invention.

[0029] Figure 2 This is a dataset of the probability of rockburst occurrence in this invention;

[0030] Figure 3 This is a dataset diagram of the rockburst intensity classification system of this invention;

[0031] Figure 4 This is a diagram of the dynamic intelligent short-term prediction system for time-delay rockbursts in deep mines, as described in this invention. Detailed Implementation

[0032] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0033] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. To facilitate understanding, the invention will now be described more fully with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0034] like Figure 1 As shown, a mine time-delay type rockburst dynamic short-term intelligent prediction system is constructed by the following steps:

[0035] Step 1: Construction of a three-dimensional microseismic field global sensing system for deep wells

[0036] Based on the distribution characteristics of mine roadways and deep rock mechanics parameters, a reconfigurable sensor array system with spatiotemporal self-correction function was established, forming a dynamic tracking microseismic sensing network topology. A three-level ring-nested layout strategy was adopted: high-density three-component intelligent microseismic sensors (sampling rate ≥10kHz) were deployed in key monitoring areas, an adaptive azimuth calibration array (including 6-DOF attitude sensors) was configured in the transition area, and broadband microseismic event acquisition nodes (1-1kHz) were arranged in the outer area. Each node integrated an edge computing module to realize the acquisition and filtering of raw microseismic waveform signals.

[0037] Step 2: Extraction of microseismic characteristic parameters

[0038] By using a tracking array of microseismic sensors during deep well mining, microseismic signal waveform data can be collected, transmitted, and stored in real time.

[0039] Time-series data containing time-dimensional feature parameters (including the number of microseismic events, microseismic event energy E, apparent stress σ, apparent volume V, volumetric potential ε, stress drop Δσ, etc., cumulative values ​​and rates of change, and moment magnitude M) are extracted from waveform and microseismic event data. w Mutation Index I).

[0040]

[0041] Temporal data from waveform and microseismic event data to extract spatial dimension temporal feature parameters (microseismic event density ρ, event density spatial-temporal fractal dimension D, and microseismic event source location (x, y, z)).

[0042]

[0043] r represents the relative size of the spatial region of the microseismic event at time t.

[0044] Step 3: Construction of the Rockburst Disaster Evolution Dataset

[0045] For the evolution process of rockburst disasters, time-series datasets of microseismic spatiotemporal characteristic parameters with rockburst intensity classification or occurrence probability labels were constructed.

[0046] ① Construct a time-series data structure with 720 timestamps (12 hours) as input and output labels for rockburst events occurring within the next 12 hours of the last input timestamp (label 1 if rockburst occurs, label 0 if rockburst does not occur). Figure 2 ;

[0047] ② When the output labels for rockbursts occurring within the next 12 hours of the last input timestamp as 1 (i.e., a rockburst occurs), it also marks the possible rockburst intensity level (0 for no rockburst, 1 for minor rockburst, 2 for moderate rockburst, and 3 for severe rockburst). For example... Figure 3 .

[0048] Step 4: Data Preprocessing

[0049] Preprocessing operations are performed on the time series dataset of microseismic spatiotemporal parameters (improved Log-Gaussian hybrid mapping normalization);

[0050] x' = log[β(x+ε)],

[0051]

[0052] Where x and x' represent the time series data of each parameter before and after preprocessing, β represents the adaptive adjustment parameter of the parameter time series data, α represents the scaling constant, and ε represents the constant to avoid the value being too small after logarithmic mapping.

[0053] An adaptive logarithmic function is used to process the time series data of each parameter, which controls the distortion effect on the feature space caused by excessive differences in the numerical increase and decrease process, and also gives a relatively large number of adaptive weights for excessively large values, but in a controlled manner.

[0054] Step 5: Deep Learning Algorithm Modeling and Training

[0055] A deep learning algorithm model, Multi-physics Spatial-Temporal Attention Networks (MSTA-Net), is adopted. The deep learning network model uses 3D-TCN as the algorithm base to extract local features of microseismic parameter time series data in the spatiotemporal dimension. The hybrid network coupled with BiGRU and Attention mechanism modules performs feature analysis on long spatiotemporal sequences. Finally, linear and nonlinear connections are used to output the probability of rockburst phenomena occurring in the next 12 hours. When the probability of rockburst phenomena occurring in the next 12 hours of the last input timestamp is greater than 50%, the intensity level of possible rockburst is output. Based on this, two dynamic time series datasets of microseismic spatiotemporal feature parameters are modeled and trained.

[0056] Data Acquisition Layer: Based on on-site mining and geological conditions, a microseismic sensor array is deployed to collect real-time data on microseismic source location (x, y, z), energy E, apparent stress σ, apparent volume V, volumetric potential ε, and moment magnitude M. w and parameters such as stress drop Δσ;

[0057] Spatiotemporal feature fusion:

[0058] Time dimension: Using time scale sliding (minimum time scale Δt = 10 min), the energy release rate ΔE / Δt and event frequency gradient dN / dt are extracted; using time window sliding method (window length T = 6 h, 12 h, 18 h and 24 h), cyclic time series data that can be input into the intelligent model are obtained.

[0059] Spatial dimension: The monitoring area is divided into a three-dimensional grid (side length ≤ 15m), and the density of microseismic events ρ per unit volume is calculated;

[0060] Early warning label generation: Combining parameters such as the degree of rockburst damage, geological conditions, and microseismic energy, a four-level rockburst intensity early warning label is constructed.

[0061] Label Y∈{0: no rockburst, 1: minor rockburst, 2: moderate rockburst, 3: severe rockburst} 2. Coupled prediction model driven by spatiotemporal attention mechanism

[0062] Model architecture: 1D CNN-BiLSTM-Attention coupled network

[0063] Input layer: Standardized multidimensional time series data matrix X(t)∈R^(N×M), where N is the time step and M is the feature dimension;

[0064] Feature extraction layer:

[0065] 1D CNN: Extracting local temporal features (kernel size = 3, number of channels = 64);

[0066] BiLSTM: Captures the spatiotemporal features of long sequences (hidden layer units = 128);

[0067] Attention layer: Calculate feature weights α i =softmax(q^T tanh(W·h) i +b)) focuses on the evolution of key disasters, integrates spatiotemporal characteristics, and regresses the degree of rockburst damage;

[0068] Output layer: The Softmax classifier outputs the probability distribution of rockburst severity P(Y|X) for the future time period Δt.

[0069] Step 6: Optimize model structure parameters

[0070] By analyzing the characteristics of the time series dataset of microseismic spatiotemporal parameters and comparing it with the bio-meta-heuristic algorithm to verify the optimization effect, the gray wolf-simulated annealing hybrid algorithm was selected to optimize the network structure parameters and model hyperparameters. An improved adaptive gray wolf algorithm was used for global search, coupled with a simulated annealing mechanism with memory effect for local fine-tuning, and finally the deep learning network model with optimal structure parameters was obtained.

[0071] Global search phase for optimizing structural parameters:

[0072]

[0073] W(k) is the adaptive weight, and X(k) is the current parameter label for the algorithm's iterative optimization. α, β, and δ represent three different label positions. K represents the convergence factor for iterative updates.

[0074] The local fine-tuning stage of optimizing structural parameters:

[0075]

[0076] P Accepted Let E(A) represent the probability of accepting the updated optimal solution, and E(B) represent the objective functions of the current solution and the new solution, respectively. T represents the current temperature during the iterative optimization of the simulated annealing algorithm. λ represents the decay constant.

[0077] Meanwhile, unlike the current research on rockburst intensity level prediction methods which rely on post-disaster summary analysis and static model prediction analysis, this step can lead to an intelligent model that can achieve real-time dynamic prediction and short-term forecasting of rockburst disasters.

[0078] Step 7: Model Functional Deconstruction Analysis

[0079] By combining statistical machine learning methods such as SHAP values ​​with deep well geological information, a dynamic short-term intelligent prediction model for rockburst disasters is functionally deconstructed. Through interpretability analysis, the black-box attribute mapping relationships of the deep learning model are made transparent, transforming it into a gray-box and even white-box model. Ultimately, under real-time microseismic monitoring conditions, an interpretable deep learning network model can be obtained through training and optimization. This model uses an end-to-end model of input data and dynamic output to evaluate the hazard level of rockburst disaster processes and accurately predict the probability of rockburst occurrence in metal mines and the intensity level of potential rockburst disasters.

[0080] Step 8: Construction of a dynamic short-term rockburst prediction system

[0081] Finally, by linking a full-domain tracking sensing system, constructing and scaling preprocessing techniques for dynamic rockburst disaster datasets based on microseismic spatiotemporal characteristic time-series data, and an intelligent prediction and forecasting model for multimodal spatiotemporal deep feature extraction, a microseismic tracking real-time monitoring-mine time-delay type rockburst dynamic intelligent short-term forecasting system was obtained. This system realizes the tracking and sensing of the spatiotemporal evolution characteristics of the microseismic field throughout the mine, the ability to extract and analyze the spatiotemporal features of rockburst disasters, and the intelligent prediction and forecasting of time-delay type rockburst dynamics. Figure 4 .

[0082] This dynamic intelligent short-term rockburst forecasting system boasts high accuracy in dynamic tracking, effectively improving data acquisition efficiency and positioning accuracy compared to traditional deployment methods. Compared to traditional intelligent rockburst intensity prediction methods, it places greater emphasis on real-time pre-disaster analysis, disaster evolution process prediction, and dynamic model forecasting for rockburst prediction. Furthermore, the system is better suited for monitoring and early warning of rockburst dynamic hazards during deep well mining operations, providing a scientific basis and technological paradigm for the subsequent perception-early warning-prevention system for time-delayed rockbursts in deep mining, encompassing data foundation, data processing, feature extraction and analysis, and forecasting.

[0083] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0084] It should be understood that the above detailed description of the technical solutions of the present invention with reference to preferred embodiments is illustrative and not restrictive. Those skilled in the art can modify the technical solutions described in the embodiments or make equivalent substitutions for some of the technical features based on reading this specification; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A time-delay type dynamic short-term intelligent prediction system for rockbursts in mines, characterized in that: The system includes the following modules: The deep well three-dimensional microseismic field global perception module, based on the distribution characteristics of mine roadways and deep rock mechanics parameters, adopts a three-level ring nested layout strategy to construct a dynamic tracking microseismic sensing network topology. High-density three-component intelligent microseismic sensors are deployed in the key monitoring area, an adaptive azimuth calibration array is configured in the transition area, and broadband microseismic event capture nodes are arranged in the outer area. Each node integrates an edge computing module for the acquisition and filtering of raw microseismic waveform signals. The microseismic feature parameter extraction module extracts time-dimensional and spatial-dimensional feature parameters by acquiring microseismic signal waveform data in real time. These parameters include the number of microseismic events, microseismic event energy, apparent volume, apparent stress, stress drop, volumetric potential, microseismic event density, and seismic source location. The specific implementation is as follows: The microseismic sensor array is tracked and deployed during the deep well mining process to collect, transmit and store microseismic signal waveform data in real time. Time-series data from waveform and microseismic event data, extracting time-dimensional feature parameters, including: number of microseismic events and energy of microseismic events. E Apparent stress σ Apparent volume V Body change Stress drop Parameter cumulative quantity and rate of change and moment magnitude Mutation index I ; ; Temporal data for extracting spatial dimension feature parameters from waveform and microseismic event data, including: microseismic event density ρ, and the spatiotemporal fractal dimension of event density. D Location of the epicenter of the microseismic event ( x, y, z ); ; r for t The relative size of the spatial region of the microseismic event at any given moment; The rockburst disaster evolution dataset construction module constructs a time series dataset of microseismic spatiotemporal characteristic parameters with rockburst intensity classification or occurrence probability labels based on the sliding time window method. The data preprocessing module uses an adaptive logarithmic function to standardize the data and optimize the feature space of time series data. The deep learning prediction module utilizes a deep learning model that couples 3D-TCN and BiGRU with an attention mechanism. It then optimizes the model's structural parameters using a gray wolf-simulated annealing hybrid algorithm to predict the probability and intensity of rockbursts within the next 12 hours. The specific implementation is as follows: Data acquisition layer: Deploy microseismic sensor arrays based on on-site mining and geological conditions to collect real-time microseismic source locations ( x, y, z ),energy E Apparent stress σ Apparent volume V Body change Moment and magnitude and stress drop parameter; Spatiotemporal feature fusion: Time dimension: Using time-scale sliding, with the smallest time-series scale Δ t =10min, extract energy release rate ΔE / Δt and event frequency gradient dN / dt; use time window sliding method, window length T =6h, 12h, 18h and 24h, to obtain the cyclic time series data of the input intelligent model; Spatial dimension: The monitoring area is divided into a three-dimensional grid with a side length ≤15m, and the microseismic event density per unit volume is calculated. ; Early warning label generation: Combining the degree of rockburst damage, geological conditions, and microseismic energy parameters, a four-level rockburst intensity early warning label is constructed: Label Y ∈{0: No rockburst, 1: Minor rockburst, 2: Moderate rockburst, 3: Severe rockburst} Coupled prediction model driven by spatiotemporal attention mechanism Model architecture: 1D CNN-BiLSTM-Attention coupled network Input layer: Standardized multidimensional time series data matrix , N For time steps, M For feature dimensions; Feature extraction layer: 1D CNN: Extracting local temporal features, kernel size = 3, number of channels = 64; BiLSTM: Captures long sequence spatiotemporal features, hidden layer units = 128; Attention layer: Calculates feature weights Focusing on the evolution of key disasters, integrating spatiotemporal characteristics, and regressing the degree of rockburst damage; Output layer: Softmax classifier outputs future Δ t Time Period Rockburst Severity Probability Distribution P ( Y | X ); The interpretability analysis module enhances the credibility of the system's prediction logic by transparently analyzing the relationship between model inputs and outputs. The dynamic short-term intelligent forecasting module enables real-time dynamic prediction and early warning of rockburst disasters based on the above modules.

2. The mine time-delay type rockburst dynamic short-term intelligent prediction system according to claim 1, characterized in that, The three-level ring-nested layout strategy in the deep well three-dimensional microseismic field global perception module includes a key monitoring area, a transition area, and an outer area. The sensor sampling rate in the key monitoring area is ≥10kHz, the sensor array in the transition area contains a 6-DOF attitude sensor, and the sampling frequency range of the broadband microseismic event capture node in the outer area is 1-1kHz.

3. The mine time-delay type rockburst dynamic short-term intelligent prediction system according to claim 1, characterized in that, The rockburst disaster evolution dataset construction module constructs a time series dataset using a sliding window method. The input is 720 timestamp data from the past 12 hours, and the output is a rockburst probability label and intensity level label for the next 12 hours. The intensity level is divided into no rockburst, minor rockburst, moderate rockburst, and severe rockburst.

4. The mine time-delay type rockburst dynamic short-term intelligent prediction system according to claim 1, characterized in that, The data preprocessing module uses an adaptive logarithmic function to standardize the data. By adjusting the scaling constant and parameters, it avoids distortion caused by excessive numerical differences and optimizes the data distribution to adapt to the input of the deep learning model.

5. The mine time-delay type rockburst dynamic short-term intelligent prediction system according to claim 1, characterized in that, The deep learning prediction module optimizes the model structure parameters using a hybrid algorithm of gray wolf and simulated annealing. The gray wolf algorithm is used for global search of structure parameters, while the simulated annealing mechanism is used for local fine-tuning optimization. The objective function is based on prediction accuracy and model generalization ability.

6. The mine time-delay type rockburst dynamic short-term intelligent prediction system according to claim 1, characterized in that, The interpretability analysis module uses the SHAP value analysis method to reveal the mapping relationship between microseismic characteristic parameters and rockburst prediction results, thereby improving the transparency and credibility of the deep learning model.

7. The mine time-delay type rockburst dynamic short-term intelligent prediction system according to claim 1, characterized in that, The dynamic short-term intelligent forecasting module combines real-time microseismic field monitoring data with deep learning prediction results to provide dynamic early warning information on rockburst disasters, including the probability of rockburst occurrence and intensity level.

Images