Coal mine rock burst dynamic early warning method based on digital twinning and edge calculation

Abstract

The invention provides a coal mine rock burst dynamic early warning method based on digital twinborn and edge calculation, and relates to the technical field of coal mine rock burst prevention and control, and the method comprises the steps: collecting working face multi-source data through a distributed Internet of Things sensing network, and constructing a three-dimensional geological digital twinborn body; calculating a ground pressure dynamic potential energy index and an edge calculation efficiency factor based on the real-time monitoring data; deploying a dynamic risk assessment model at an edge node, and performing risk dynamic grading in combination with a digital twinborn simulation result; optimizing an edge computing task allocation strategy and constructing a multi-algorithm collaborative early warning mechanism; multi-node early warning cross validation is realized through spatio-temporal correlation analysis; and triggering a grading early warning signal according to the dynamic threshold value and synchronously updating the digital twinborn parameters. According to the invention, through deep fusion of digital twinning and edge calculation, real-time processing and dynamic modeling of multi-source data are realized, and the problems of high delay and data islands of a traditional system are solved; a hybrid algorithm architecture and dynamic parameter optimization are utilized to improve the generalization ability of the model, and the false alarm rate is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine rock burst prevention and control, and in particular to a coal mine rock burst dynamic early warning method based on digital twin and edge computing. Background Art

[0002] In the field of coal mine rock burst monitoring and early warning, existing technical solutions generally face the dual challenges of real-time performance and dynamic adaptability. Traditional systems mostly rely on centralized data processing architectures, and monitoring data needs to be transmitted over long distances to ground servers for analysis, resulting in a significant increase in data processing delays. Especially under the influence of complex electromagnetic environments and network fluctuations underground, the data transmission bandwidth is limited, making it difficult to meet the timeliness requirements of rock burst early warning for millisecond-level responses. In addition, existing methods lack the ability to integrate multi-source heterogeneous data. Geological structure data, microseismic signals, stress field distribution, and equipment status information are often processed independently. There is a lack of unified dynamic modeling methods, which makes it impossible for risk assessment models to fully reflect the true state of the working face.

[0003] Current early warning models often use fixed thresholds or periodically updated static parameters, such as preset stress thresholds based on historical data or periodically revised rock mechanics parameters. However, the geological conditions of coal mine working faces continuously change as mining progresses, and key parameters such as gas concentration and rock formation stress exhibit dynamic fluctuations. Static models struggle to capture these changes in real time, leading to delayed risk assessments and accumulated bias. Existing systems, especially in the event of sudden fault activation or rock formation fractures, often misjudge or miss warnings due to infrequent model updates.

[0004] At the algorithmic level, single machine learning models (such as support vector machines or random forests) have limited generalization capabilities. When faced with complex and changing underground monitoring data, they are susceptible to noise interference or changes in feature correlation, leading to increased false alarm rates. Furthermore, traditional algorithms lack the ability to dynamically adapt to edge computing resources and have rigid task allocation strategies. This often results in some nodes being overloaded while others are idle, and overall computing performance is underutilized.

[0005] Existing early warning mechanisms lack the ability to coordinate multiple edge nodes. Each node operates independently and lacks a data cross-validation mechanism, leading to local misjudgments that can directly impact global warning results. Furthermore, the warning level classification standards are overly general, typically setting only a single threshold to trigger a fixed response measure. This makes it impossible to dynamically adjust emergency strategies based on risk levels, making it difficult to achieve precise tiered management and control. These issues severely restrict the reliability and practicality of coal mine rock burst warning systems, necessitating a comprehensive solution that integrates dynamic modeling, real-time optimization, and intelligent decision-making. Summary of the Invention

[0006] In order to solve the technical problems in the existing technology such as high latency and lack of real-time performance caused by centralized data processing architecture, data islands formed by the lack of multi-source heterogeneous data fusion, high false alarm rate caused by the difficulty of static parameter models to adapt to dynamic geological changes, insufficient generalization ability of single machine learning algorithms, node load imbalance caused by rigid edge computing resource allocation, local misjudgment spread caused by lack of collaborative verification of multiple edge nodes, and rough warning level division that cannot achieve accurate emergency response, the present invention provides a dynamic early warning method for coal mine rock burst based on digital twins and edge computing.

[0007] The technical solutions provided by the present invention are as follows:

[0008] The present invention provides a dynamic early warning method for coal mine rock burst based on digital twin and edge computing, including:

[0009] S1. Collect working face geological structure data, microseismic monitoring data, stress field distribution data, and equipment operation status data through a distributed IoT sensor network to build a three-dimensional geological digital twin.

[0010] S2. Calculate two key parameters: the dynamic potential energy index (DPEI) and the edge computing efficiency factor (ECEF) based on real-time monitoring data.

[0011] S3. Deploy a dynamic risk assessment model on edge computing nodes and conduct dynamic assessment of rock burst risk based on the real-time simulation results of the digital twin.

[0012] S4, dynamically optimize the task allocation strategy of edge computing nodes based on ECEF parameters and establish a multi-algorithm collaborative early warning mechanism;

[0013] S5. Use a hybrid algorithm architecture to integrate support vector machines, random forests, and deep learning algorithms to form a composite early warning decision model;

[0014] S6. Establish a multi-edge node collaborative computing framework based on spatiotemporal correlation analysis to achieve cross-validation of warning results;

[0015] S7. When the warning level exceeds the set threshold, a graded warning signal is triggered through the low-latency communication network, and the digital twin model parameters are updated synchronously.

[0016] Furthermore, the S2 further includes:

[0017] Based on real-time monitoring data, the dynamic potential energy index (DPEI) of the ground pressure is calculated using the following formula:

[0018]

[0019] Among them, σ i is the real-time stress value of the monitoring point; ε i is the corresponding deformation rate; T is the geological structure complexity coefficient; C g is the gas concentration; P w is the current working face advancement speed; P0 is the benchmark advancement speed.

[0020] Furthermore, the S3 further includes:

[0021] The improved Mohr-Coulomb criterion is used for risk assessment, which is specifically achieved through the following shear stress criterion formula:

[0022] τ c =σ n tanφ+c(1-e -k·t )

[0023] Among them, τ c is the critical shear stress; σ n is the normal stress; φ is the internal friction angle; c is the cohesion; k is the time effect coefficient; t is the continuous loading time.

[0024] Furthermore, the S2 further includes:

[0025] The edge computing efficiency factor ECEF is calculated using the following formula:

[0026]

[0027] Among them, α is the communication quality correction coefficient (0.6-0.8); β is the node load balancing coefficient (1.2-1.5); L max is the maximum computation delay; L avg is the average calculation delay; D avg is the current data transmission volume; D max The maximum theoretical transmission capacity.

[0028] Furthermore, the S5 specifically includes:

[0029] The weights of each algorithm are optimized by genetic algorithm, and the optimization process adopts the following fitness function:

[0030] Fitness=ω1·AUC+ω2·(1-FPR)+ω3·TTA

[0031] Among them, AUC is the area under the ROC curve of the model; FPR is the false alarm rate; TTA is the warning response time; ω1, ω2, and ω3 are dynamic weight coefficients optimized by genetic algorithm.

[0032] Further, the parameter update in S7 specifically includes:

[0033] When the deviation of the monitoring data exceeds 5% or the time interval reaches 10 minutes, the update step is automatically triggered, and this step uses the Kalman filtering algorithm for data fusion processing.

[0034] Further, S1 specifically includes three-level preprocessing steps:

[0035] The first preprocessing step uses wavelet transform for signal denoising; the second preprocessing step uses the sliding window algorithm for outlier detection; the third preprocessing step uses the improved DBSCAN clustering algorithm to achieve data feature stratification.

[0036] Further, the task assignment strategy of S4 specifically includes:

[0037] Construct a dynamic linear programming model and perform optimization calculations through the following objective function:

[0038]

[0039] where c i is the node calculation cost; x i is the task assignment decision variable; d j is the task deadline; t j is the estimated completion time; λ is the delay penalty coefficient.

[0040] Further, the threshold setting in S7 includes:

[0041] The step of determining the reference value based on the Weibull distribution fitting of historical accident data and dynamically adjusting it in combination with real-time ECEF parameters.

[0042] Further, the hierarchical warning in S7 includes:

[0043] The first response step: When DPEI > 0.8, trigger a red warning and initiate an emergency evacuation; the second response step: When 0.6 < DPEI ≤ 0.8, trigger an orange warning and limit the number of operating personnel; the third response step: When DPEI ≤ 0.6, maintain a yellow warning and increase the monitoring frequency

[0044] The beneficial effects brought by the technical solution provided by the present invention at least include:

[0045] (1) In the present invention, by constructing a collaborative architecture of three-dimensional geological digital twins and edge computing nodes, real-time fusion and localized processing of multi-source heterogeneous data are achieved. Digital twin technology dynamically maps geological structure, stress field distribution, and equipment status data into high-precision simulation models. Combined with the local processing of monitoring data by edge computing nodes, it effectively reduces data transmission delays and improves early warning response speed. At the same time, the distributed deployment mode of edge nodes optimizes the allocation of computing resources, avoids the bandwidth bottleneck of centralized architecture, significantly improves the system's adaptability to complex underground environments, and solves the problems of insufficient real-time performance and data silos in traditional systems.

[0046] (2) In the present invention, a hybrid algorithm architecture and dynamic parameter optimization mechanism are adopted to enhance the generalization ability and dynamic adaptability of the early warning model. By integrating the advantages of support vector machines, random forests, and deep learning algorithms, combined with genetic algorithms to dynamically optimize weight distribution, the sensitivity of a single model to noisy data is reduced, and the accuracy of early warnings under complex working conditions is improved. The dynamic risk assessment model introduces a time effect factor and a real-time geological parameter update mechanism, enabling the model to automatically track changes in rock stress and mining progress, solving the problem of assessment bias caused by the difficulty of static models in capturing dynamic geological changes.

[0047] (3) In the present invention, based on a multi-edge node collaborative computing framework and a hierarchical early warning mechanism, accurate risk identification and emergency response optimization are achieved. Spatiotemporal correlation analysis supports cross-validation of multi-node early warning results, and combined with digital twin simulation data, local misjudgments are corrected to improve the reliability of global early warnings. The hierarchical early warning signal triggers differentiated emergency measures based on the dynamic potential energy index threshold. By adaptively adjusting the threshold and matching the risk level with the multi-level response strategy, the problem of the extensive response of the traditional early warning mechanism is solved, and precise control of the entire process from risk warning to emergency response is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 A flow chart of a dynamic early warning method for coal mine rock burst based on digital twin and edge computing provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0051] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0052] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0053] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0054] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0055] Reference Manual Figure 1 , which shows a flow chart of the dynamic early warning method for coal mine rock burst based on digital twin and edge computing provided by an embodiment of the present invention.

[0056] The embodiment of the present invention provides a dynamic early warning method for coal mine rock burst based on digital twin and edge computing. The processing flow may include the following steps:

[0057] S1. Collect working face geological structure data, microseismic monitoring data, stress field distribution data and equipment operation status data through a distributed Internet of Things sensor network to construct a three-dimensional geological digital twin.

[0058] It should be noted that multi-dimensional data is collected through a distributed Internet of Things sensor network deployed on the coal mine working face. The sensor network consists of microseismic sensors, stress sensors, gas concentration sensors and equipment status monitoring devices.

[0059] Furthermore, after data collection, three levels of preprocessing are performed: the first level of preprocessing uses discrete wavelet transform (DWT) to decompose and reconstruct the original signal and filter out high-frequency noise; the second level of preprocessing segments the time series data through a sliding window algorithm, calculates the standard deviation and mean of each segment of data, and if a data point in a window deviates from the mean by more than three times the standard deviation, it is marked as an outlier and eliminated; the third level of preprocessing applies an improved density-based spatial clustering of applications with noise (DBSCAN) algorithm, which clusters data with similar characteristics into different levels by dynamically adjusting the neighborhood radius parameter ε and the minimum number of samples MinPts.

[0060] In the third stage preprocessing in one possible implementation, the DBSCAN clustering algorithm is modified to use the following adaptive neighborhood radius formula:

[0061]

[0062] Among them, ε t is the neighborhood radius at the current moment, ε t-1 is the radius at the previous moment, and Δρ is the rate of change of data density.

[0063] When Δρ>20%, ε t Reduced to 80% of its original value;

[0064] When Δρ<-15%, ε t Expanded to 120% of its original value.

[0065] The clustering validity is verified by the Silhouette Coefficient. If the coefficient is lower than 0.5, the manual intervention process is triggered.

[0066] Furthermore, the improved DBSCAN algorithm sets the initial neighborhood radius to 1.5 times the average distance between data points and dynamically adjusts based on the density of new data: if the density of an area increases by more than 20%, the radius decreases to 80% of the original value; if the density decreases by more than 15%, the radius increases to 120% of the original value. The minimum number of samples adjusts synchronously with the radius to ensure clustering stability.

[0067] The pre-processed data is input into the digital twin modeling engine, and the geological structure parameters and historical data are combined to build a three-dimensional geological digital twin. The model maps the geological structure of the working face, stress field distribution and equipment operation status in real time.

[0068] S2. Calculate two key parameters, ground pressure dynamic potential energy index and edge computing efficiency factor, based on real-time monitoring data.

[0069] Specifically, for the calculation of DPEI, the real-time stress value σ of each monitoring point is first extracted from the preprocessed data. i and the corresponding deformation rate ε i , based on the formula

[0070]

[0071] Comprehensive calculation is performed, in which the geological structure complexity coefficient T is determined by analyzing the fault density and rock layer inclination of the working face, and the gas concentration C g The current working face advancement speed P is obtained by real-time monitoring of the gas sensor. w The benchmark advancing speed P0 is dynamically updated through the coal mining machine operation log.

[0072] Furthermore, the geological complexity coefficient T is calculated by combining the fault density and rock formation dip angle of the working face. In practice, fault density is quantified by the total length of fault lines per unit area, and rock formation dip angle is averaged based on borehole exploration data and 3D laser scanning results. A dynamic coefficient is generated using a weighted approach, with fault density accounting for 70% and the sine value of rock formation dip accounting for 30%. This coefficient is updated once per shift based on the latest geological exploration data.

[0073] It should also be noted that for the calculation of ECEF, the maximum calculation delay L of the edge computing node is first monitored. max , average computation delay L avg , Current data transmission volume D avg And the maximum theoretical transmission capacity D max , combined with the preset communication quality correction coefficient α (value 0.6-0.8) and the node load balancing coefficient β (value 1.2-1.5), through the formula

[0074]

[0075] Dynamically evaluate the computing performance of edge nodes.

[0076] It should be noted that the communication quality correction coefficient α ranges from 0.6 to 0.8, and its implementation needs to be dynamically adjusted according to the wireless signal strength and data packet loss rate: when the signal strength is higher than -70dBm and the packet loss rate is lower than 2%, the upper limit value is 0.8; when the signal strength is lower than -90dBm or the packet loss rate exceeds 10%, the lower limit value is 0.6; the intermediate state is calculated by interpolation based on the linear relationship between signal strength and packet loss rate.

[0077] The node load balancing coefficient β is adjusted according to the real-time CPU utilization differences of each edge node. If the standard deviation of utilization between nodes exceeds 15%, the coefficient is increased to 1.5 to enhance load balancing.

[0078] S3. Deploy a dynamic risk assessment model at the edge computing node and perform a dynamic assessment of rock burst risk based on the real-time simulation results of the digital twin.

[0079] Specifically, this is achieved by deploying a dynamic risk assessment model at the edge computing node, which combines the improved Mohr-Coulomb criterion with the digital twin simulation results. The improved Mohr-Coulomb criterion adopts the shear stress criterion formula

[0080] τ c =σ n tanφ+c(1-e -k·t )

[0081] Where the normal stress σ n The internal friction angle φ is obtained from the rock mass mechanics parameter library provided by the digital twin, the cohesion c is calibrated by laboratory rock sample testing, the time effect coefficient k is determined by fitting historical rock mass creep data, and the continuous loading time t is obtained by the cumulative duration of microseismic events. When the model is running, the critical shear stress τ is calculated in real time. c and the actual shear stress τ actual In contrast, if the actual value exceeds the critical value, a risk level increase signal is triggered.

[0082] It should be noted that the actual shear stress τ actual and critical shear stress τ c The judgment conditions are:

[0083] If τ actual ≥1.2τ c , it is judged as extremely high risk and directly triggers a red alert;

[0084] If 1.0τ c ≤τ actual <1.2τ c , determined to be high risk, and requires verification in conjunction with digital twin simulation;

[0085] If τ actual <1.0τ c , but there are more than 2 τ in the three adjacent monitoring points actual ≥0.9τ c , it was determined to be a potential risk and an orange alert was activated.

[0086] Furthermore, the time-effect coefficient k in the improved Mohr-Coulomb criterion must be determined through rock mass creep experiments. Triaxial creep tests are conducted on different rock formation types (such as sandstone and mudstone). The slope of the second phase of the creep curve is extracted as a baseline coefficient value and stored in the rock mass parameter library of the digital twin. The calculation of the continuous loading time t requires the accumulation of the duration of the stress fluctuation during the microseismic event exceeding 80% of the static rock stress.

[0087] S4. Dynamically optimize the task allocation strategy of edge computing nodes based on ECEF parameters and establish a multi-algorithm collaborative early warning mechanism.

[0088] Specifically, the multi-algorithm collaborative early warning is achieved by dynamically optimizing the task allocation strategy of edge computing nodes. The task allocation strategy adopts a dynamic linear programming model, and the objective function is

[0089]

[0090] The node computation cost c i According to the pre-calibration of node hardware performance and energy consumption indicators, the task allocation decision variable x i It is a 0-1 variable indicating whether the task is assigned to the i-th node and the task deadline d j and the estimated completion time t j By dynamically estimating task complexity and node load, the delay penalty coefficient λ is dynamically adjusted according to the warning level. The optimization process uses a branch-and-bound algorithm to solve the problem, and the optimal allocation solution is sent to the edge nodes for execution.

[0091] Furthermore, the delay penalty coefficient λ in the dynamic linear programming model is directly linked to the alert level: During a red alert, the coefficient is set to 5.0 to strictly limit task delays. During an orange alert, it is reduced to 3.0, and during a yellow alert, it is further reduced to 1.5. The estimated completion time calculation requires the node's real-time load factor, which is a weighted average of CPU and memory utilization, with weights of 60% and 40%, respectively.

[0092] S5. Use a hybrid algorithm architecture to integrate support vector machine, random forest and deep learning algorithms to form a composite early warning decision model.

[0093] Specifically, the support vector machine (SVM), random forest (RF) and deep learning algorithms are integrated through a hybrid algorithm architecture. The weight distribution of each algorithm is optimized by the genetic algorithm, and the fitness function is

[0094] Fitness=ω1·AUC+ω2·(1-FPR)+ω3·TTAS

[0095] Where AUC is the area under the receiver operating characteristic curve (ROC), FPR is the false alarm rate, and TTA is the warning response time. The dynamic weight coefficients ω1, ω2, and ω3 are iteratively optimized through selection, crossover, and mutation operations using a genetic algorithm. The resulting composite warning decision model outputs a comprehensive risk score and classifies warning levels based on the score.

[0096] Specifically, during the optimization process of the genetic algorithm, it is necessary to constrain the sum of the weight coefficients to be 1, and each single weight should not be less than 0.2. The iteration termination condition is that the fitness improvement amplitude is less than 0.1% for ten consecutive generations or the total number of iterations reaches 200 times. The crossover operation adopts simulated binary crossover, and the mutation probability is fixed at 5% to ensure the optimization efficiency and rationality.

[0097] It should be noted that the dynamic adjustment rules for the weights of each algorithm are as follows:

[0098] When the signal-to-noise ratio (SNR) < 10 dB, the weight ω3 of the deep learning algorithm is increased to 0.5, and the weight ω1 of the support vector machine is decreased to 0.2;

[0099] When the feature dimension exceeds 50 dimensions, the weight ω2 of the random forest is increased to 0.6;

[0100] If the warning results are inconsistent for three consecutive times, the genetic algorithm re-initializes the population and optimizes the weights.

[0101] S6. Establish a multi-edge node collaborative computing framework based on spatio-temporal correlation analysis to achieve cross-verification of the warning results.

[0102] Specifically, a multi-edge node collaborative computing framework is constructed based on spatio-temporal correlation analysis. After each edge node independently calculates the local risk index, the data trends in adjacent regions are analyzed through the spatio-temporal correlation matrix to cross-verify the conflicting warning results. If more than half of the nodes have the same risk assessment results for the same region, it is determined as a valid warning; if there are significant differences, the digital twin is triggered to re-simulate and the manual review process is started.

[0103] S7. When the warning level exceeds the set threshold, trigger a graded warning signal through a low-latency communication network and synchronously update the parameters of the digital twin model.

[0104] It should be noted that this step includes two parts: triggering the graded warning and updating the model parameters.

[0105] Specifically, the warning grading is based on the DPEI threshold: when DPEI > 0.8, a red warning is triggered, and the audible and visual alarm device is automatically activated and an emergency evacuation instruction is sent to the underground terminal; when 0.6 < DPEI ≤ 0.8, an orange warning is triggered, non-essential operating personnel are restricted from entering high-risk areas, and the advancing speed of the coal mining machine is reduced; when DPEI ≤ 0.6, the yellow warning state is maintained, and the monitoring frequency is increased to once per minute.

[0106] In a possible implementation, the following auxiliary judgment conditions can be added:

[0107] The red warning needs to meet all the following conditions:

[0108] [[ID=]7]]DPEI > 0.8;

[0109] The gas concentration increases by 15% or more compared to the previous ten minutes;

[0110] The frequency of microseismic events is ≥5 times / minute.

[0111] An orange alert must meet any of the following conditions:

[0112] DPEI∈(0.6,0.8] and ECEF<0.7;

[0113] DEPI≤0.6 but τ actual ≥0.95τ c Lasts for 5 minutes.

[0114] Threshold setting involves fitting historical accident data with a Weibull distribution to determine a baseline value, and dynamically adjusting it based on real-time ECEF parameters. Fitting the Weibull distribution to historical accident data requires extracting a parameter sequence three hours before and after the accident, solving for shape and scale parameters using maximum likelihood estimation, and dynamically adjusting the threshold based on the fitting results, allowing it to decrease as edge computing performance improves.

[0115] It's also important to note that model parameter updates utilize a dual-trigger mechanism: an update is triggered immediately when the deviation between real-time monitoring data and the digital twin's predicted value exceeds 5%, or a periodic update is performed every 10 minutes. During the update process, a Kalman filter algorithm is used to fuse sensor data with simulation results to correct the digital twin's stress field distribution and device state parameters.

[0116] It should be noted that the Kalman filter data fusion formula is expanded to:

[0117]

[0118] Among them, Z k is the sensor observation value, H is the observation matrix, K k is the Kalman gain. During the update process, if the residual (σ is the historical residual standard deviation), it is determined that the sensor is abnormal, the update is suspended and the fault diagnosis program is started

[0119] Furthermore, the determination of monitoring data deviation exceeding 5% is based on the relative error between the real-time monitoring value and the digital twin's predicted value. The error denominator uses the parameter dimension benchmark value (e.g., stress is based on 10 MPa) to protect low-level parameters. When the error exceeds 5%, it is determined to be a significant deviation and triggers a model update.

[0120] When the warning results of the support vector machine, random forest and deep learning algorithms are inconsistent, the results supported by both algorithms are given priority; if all three are different, the decision is made by voting based on the historical accuracy weights (30% for support vector machine, 40% for random forest, and 30% for deep learning), and the digital twin simulation is triggered to verify the final result.

[0121] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0122] (1) In the present invention, by constructing a collaborative architecture of three-dimensional geological digital twins and edge computing nodes, real-time fusion and localized processing of multi-source heterogeneous data are achieved. Digital twin technology dynamically maps geological structure, stress field distribution, and equipment status data into high-precision simulation models. Combined with the local processing of monitoring data by edge computing nodes, it effectively reduces data transmission delays and improves early warning response speed. At the same time, the distributed deployment mode of edge nodes optimizes the allocation of computing resources, avoids the bandwidth bottleneck of centralized architecture, significantly improves the system's adaptability to complex underground environments, and solves the problems of insufficient real-time performance and data silos in traditional systems.

[0123] (2) In the present invention, a hybrid algorithm architecture and dynamic parameter optimization mechanism are adopted to enhance the generalization ability and dynamic adaptability of the early warning model. By integrating the advantages of support vector machines, random forests, and deep learning algorithms, combined with genetic algorithms to dynamically optimize weight distribution, the sensitivity of a single model to noisy data is reduced, and the accuracy of early warnings under complex working conditions is improved. The dynamic risk assessment model introduces a time effect factor and a real-time geological parameter update mechanism, enabling the model to automatically track changes in rock stress and mining progress, solving the problem of assessment bias caused by the difficulty of static models in capturing dynamic geological changes.

[0124] (3) In the present invention, based on a multi-edge node collaborative computing framework and a hierarchical early warning mechanism, accurate risk identification and emergency response optimization are achieved. Spatiotemporal correlation analysis supports cross-validation of multi-node early warning results, and combined with digital twin simulation data, local misjudgments are corrected to improve the reliability of global early warnings. The hierarchical early warning signal triggers differentiated emergency measures based on the dynamic potential energy index threshold. By adaptively adjusting the threshold and matching the risk level with the multi-level response strategy, the problem of the extensive response of the traditional early warning mechanism is solved, and precise control of the entire process from risk warning to emergency response is achieved.

[0125] The above content is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0126] There are a few points to note:

[0127] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention. Other structures may refer to conventional designs.

[0128] (2) For the sake of clarity, the thickness of layers or regions in the drawings used to describe the embodiments of the present invention are exaggerated or reduced, that is, these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region, or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or intervening elements may be present.

[0129] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to form new embodiments.

[0130] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A dynamic early warning method for coal mine rock burst based on digital twins and edge computing, characterized by: Including: S1. Collect geological structure data, microseismic monitoring data, stress field distribution data, and equipment operation status data of the working face through a distributed Internet of Things sensing network, and construct a three-dimensional geological digital twin; S2. Calculate two key parameters, the ground pressure dynamic potential energy index DPEI and the edge computing efficiency factor ECEF, based on real-time monitoring data; S3. Deploy a dynamic risk assessment model at the edge computing node, and conduct dynamic assessment of rock burst risks in combination with the real-time simulation results of the digital twin; S4. Dynamically optimize the task allocation strategy of the edge computing node according to the ECEF parameter, and establish a multi-algorithm collaborative early warning mechanism; S5. Adopt a hybrid algorithm architecture to fuse support vector machine, random forest, and deep learning algorithms to form a composite early warning decision model; S6. Establish a multi-edge node collaborative computing framework based on spatio-temporal correlation analysis to achieve cross-verification of early warning results; S7. When the early warning level exceeds the set threshold, trigger a hierarchical early warning signal through a low-latency communication network, and synchronously update the digital twin model parameters.

2. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1 is characterized in that: The S2 further includes: Based on real-time monitoring data, calculate the ground pressure dynamic potential energy index DPEI through the following formula: Among them, σ i is the real-time stress value of the monitoring point; ε i is the corresponding deformation rate; T is the geological structure complexity coefficient; C g is the gas concentration; P w is the current working face advancement speed; P0 is the benchmark advancement speed.

3. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1 is characterized in that: The S3 further includes: Apply the improved Mohr-Coulomb criterion for risk assessment, which is specifically implemented through the following shear stress criterion formula: t c =s n tanφ+c(1-e -k·t ) Among them, τ c is the critical shear stress; σ n is the normal stress; φ is the internal friction angle; c is the cohesion; k is the time effect coefficient; t is the continuous loading time.

4. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1 is characterized in that: The S ​ Among them, α is the communication quality correction coefficient (0.6-0.8); β is the node load balancing coefficient (1.2-1.5); L max is the maximum computation delay; L avg is the average calculation delay; D avg is the current data transmission volume; D max The maximum theoretical transmission capacity.

5. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1 is characterized in that: ​ ​ ​ ​ 6. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1 is characterized in that: ​ ​ 7. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1 is characterized in that: ​ ​ 8. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1 is characterized in that: ​ ​ Among them, c i Calculate the cost for the node; x i Assign decision variables to tasks; d j is the task deadline; t j is the estimated completion time; λ is the delay penalty coefficient.

9. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1 is characterized in that: ​ ​ 10. The coal mine rock burst dynamic early warning method based on digital twin and edge computing according to claim 1, characterized in that: ​ ​

Images