A coal mine underground roadway and working face support monitoring method

Abstract

This invention discloses a method for monitoring the support of underground roadways and working faces in coal mines. The method includes: constructing a distributed optical fiber and multi-dimensional electromagnetic radiation collaborative sensing network to simultaneously acquire vibration and broadband electromagnetic signals along the entire line; using the CEEMDAN algorithm combined with the permutation entropy discrimination criterion and an improved wavelet threshold function to adaptively eliminate strong noise in the construction environment and reconstruct the effective energy of micro-seismic events; extracting a combined acoustic and magnetic feature vector containing micro-seismic b-values, moment magnitudes, and electromagnetic dominant frequency redshifts, and inverting the support damage factor through a multi-physics coupling model; constructing an NGO-CNN-BiGRU-Attention deep learning model to predict the instability probability, and introducing D-S evidence theory for decision-level fusion. This invention effectively solves the problems of low signal-to-noise ratio and delayed identification of disaster precursors in deep excavation sites, achieving all-weather, high-precision intelligent early warning of support structure stability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal mining, in particular to a coal mine underground roadway and working face support monitoring method. BACKGROUND

[0002] Underground engineering support monitoring is a crucial safety guarantee for deep resource development, large-scale water conservancy and hydropower construction, and urban deep underground space utilization. With the continuous increase of excavation depth, the frequency and intensity of dynamic disasters such as brittle surrounding rock rockburst, large deformation of roadway, and cracking of support lining induced by high ground stress significantly increase. In order to evaluate the safety of the support structure in real time, researchers have proposed various monitoring methods aimed at capturing physical information during rock failure to achieve early warning of disasters.

[0003] Current monitoring methods mainly fall into two categories: point monitoring and distributed monitoring. Traditional point monitoring methods, such as using anchor force meters, vibrating wire strain gauges, point seismometers, or microseismic probes, mainly acquire local physical quantities by installing sensors at specific locations. For example, the widely used microseismic monitoring technology in coal mine safety management captures elastic waves generated by rock micro-cracks through sensor array arrangement. However, point monitoring technology has significant limitations in the extremely complex environment of underground engineering: first, the monitoring range of point sensors is extremely limited, and for kilometers-long roadways or tunnels, it is difficult to achieve full coverage, leaving a large number of monitoring blind spots; second, the installation and wiring of traditional sensors are extremely complex, and in the environment of ultra-deep high ground temperature, high humidity, and strong ground electric interference, sensors are easily damaged, leading to the failure of the monitoring system.

[0004] In response to the shortcomings of point monitoring, distributed fiber optic sensing technology (DFOS) emerged. Patent EP2630519A2 discloses a method for monitoring underground transmission pipelines using distributed acoustic sensing (DAS) technology, which uses optical fiber as the sensing medium to obtain vibration information along the entire length of the line by demodulating the phase change of Rayleigh echoes. Distributed fiber sensing has the advantages of anti-electromagnetic interference, intrinsic safety, wide monitoring range, and high spatial resolution, and can convert ordinary optical cables into a "continuous listener" with thousands of sensing points. However, in the underground engineering excavation construction site, this technology faces severe challenges. The booming of tunneling machines, blast shock waves, heavy vehicle driving, and other strong noise signals generated during construction often exceed the real surrounding rock micro-crack signals (microseismic) by several orders of magnitude. Traditional linear filtering techniques cannot extract microseismic features from such complex backgrounds, resulting in a significant decrease in the accuracy of distributed fiber monitoring in rock failure positioning and energy assessment.

[0005] Furthermore, monitoring a single vibration signal often fails to comprehensively reflect the multi-field evolution characteristics of rock masses. Rock mechanics studies show that during the process of rock mass deformation under pressure until instability and fracturing, not only is elastic wave release accompanied by the generation of significant electromagnetic radiation (EMR) signals due to charge separation, triboelectric charging, and piezoelectric effects. Electromagnetic signals propagate rapidly and are extremely sensitive to changes in the stress state of rock masses, especially during the "quiet period" before macroscopic failure, where electromagnetic radiation often exhibits obvious abnormal peaks, providing excellent early warning indications. Although Chinese patent CN115291281B proposes a real-time microseismic magnitude calculation method based on deep learning, its algorithm architecture is simplistic and fails to deeply couple electromagnetic precursor information with vibration information, making it difficult to simultaneously achieve timely and spatial accuracy in early warning.

[0006] The core contradictions in current underground engineering support monitoring can be summarized as follows: the contradiction between the requirement for full coverage of distributed sensing and the strong background noise interference at the excavation site; the contradiction between the sensitivity of microseismic event capture and the risk of false alarms caused by large-scale blasting disturbances; and the contradiction between the richness of multi-physics field information and the efficiency of data fusion and real-time decision-making. These contradictions limit the widespread application of existing technologies in ultra-deep, high-hazard-risk underground engineering. Therefore, there is an urgent need for a new support monitoring method that can efficiently denoise, deeply fuse seismic and magnetic information, and provide intelligent early warning based on dynamic stress models. Summary of the Invention

[0007] The purpose of this invention is to provide a method for monitoring the support of underground roadways and working faces in coal mines, so as to solve the technical problems existing in the background art.

[0008] To achieve the above objectives,

[0009] A method for monitoring the support of underground roadways and working faces in coal mines, the method comprising the following steps:

[0010] Step 1: Lay flame-retardant and anti-static distributed optical fiber sensing cables along the roadway axis or in the roadway perimeter of the support structure in the coal mine's longwall mining roadway, tunneling roadway, or fully mechanized mining face. Construct a distributed vibration sensing network based on phase-sensitive optical time-domain reflectometry (PTZ) technology to acquire real-time vibration time-series signals of the entire coal and rock mass and support structure. Simultaneously, install intrinsically safe multi-dimensional electromagnetic radiation detection antennas in the advanced support area and stress concentration area of ​​the mining face to synchronously acquire broadband electromagnetic radiation pulse signals generated by the coal and rock mass under mining stress in real time.

[0011] Step 2: The collected distributed vibration signal is nonlinearly and nonstationarily decomposed using the CEEMDAN algorithm to generate a series of intrinsic mode function components. By calculating the permutation entropy value of each intrinsic mode function component, a component attribute identification matrix is ​​constructed to identify and remove high-entropy mechanical noise components affected by coal mining machine cutting, scraper conveyor operation, and hydraulic support movement, while retaining low-entropy dominant components containing information on coal and rock micro-fracture characteristics. For the retained components, an improved wavelet threshold function is introduced for further noise reduction processing to reconstruct the useful signal of coal and rock micro-vibration.

[0012] Step 3: Extract the spatial location coordinates, energy release rate, dominant frequency, and indicators reflecting the evolution trend of rockbursts from the preprocessed distributed vibration signals. Value sequence; synchronously extract electromagnetic pulse intensity, pulse count rate and main frequency redshift features of electromagnetic waves from electromagnetic radiation signals, and establish a joint feature vector matrix covering acoustic and magnetic features;

[0013] Step 4: Using a pre-defined coupled mathematical model of mining stress-microseismic activity-gas-electromagnetic radiation, input the joint feature vector extracted in Step 3 into this coupled model to invert and obtain the dynamic stress state and damage factor of the coal mine roadway support structure and coal-rock interface. ;

[0014] Step 5: Construct a deep learning fusion early warning model based on the NGO-CNN-BiGRU-Attention architecture. Use multi-physics feature vectors and inverted damage factors as inputs to predict the instability probability and rockburst hazard level of the support structure in the future. Use the DS evidence theory based on Bray-Curtis dissimilarity improvement to fuse the early warning conclusions from fiber optic sensing and electromagnetic sensing at the decision layer, and output the final early warning signal and targeted dynamic support adjustment suggestions.

[0015] Furthermore, in step two, the CEEMDAN algorithm is used to analyze the vibration signal. The specific mathematical implementation process for decomposition is as follows: Define the operator The first generated by empirical mode decomposition Each intrinsic mode component Let i be a standard Gaussian white noise sequence with zero mean and constant variance for the i-th experiment, where i represents the experiment number. The noise coefficient is added when processing the k-th residual signal, and I is the total number of experiments; first, a signal affected by mining noise is constructed. Empirical mode decomposition is performed on it to obtain the first intrinsic mode component. Mean:

[0016] ,

[0017] Calculate the first residual term Then, an adaptive noise component is added to the residual term, and the 1st term is iteratively calculated. One IMF component:

[0018] ,

[0019] The complete expansion of the signal is finally obtained until the residual signal no longer satisfies the decomposition conditions. .

[0020] Furthermore, the component selection in step two employs the permutation entropy criterion, and its specific steps include: reconstructing the phase space for each IMF component to obtain a matrix. Its row vectors contain the embedding dimension. and delay time The time series segment; the elements in the row vector are sorted in ascending order, and the probability of each sorting pattern is calculated. The permutation entropy of this component is calculated using the Shannon entropy formula. : Set the permutation entropy threshold It is 0.6, when the component is If the noise is determined to be mechanical vibration noise generated by the coal mining machine or conveyor, it will be removed.

[0021] Furthermore, in step three, when extracting microseismic features, a mechanism reflecting the energy release pattern of coal and rock is introduced. The value calculation formula is based on the statistical law of the magnitude-frequency relationship: ,in, The magnitude is [missing information]. For earthquakes with a magnitude greater than or equal to The cumulative number of microseismic events This is a parameter representing the regional activity level; calculated using a sliding time window. The real-time evolution curve of the value, when When the value shows a significant and continuous decline, it is determined that the working face or roadway is threatened by rock burst.

[0022] Furthermore, in the coupled mathematical model of mining stress-microseismic-electromagnetic radiation in step four, the composite damage factor... The calculation formula is: ,in, This represents the total number of electromagnetic radiation pulses detected per unit time. This represents the maximum instantaneous electromagnetic pulse value at which the coal body reaches its peak strength under the geological conditions of this coal seam.

[0023] Furthermore, the NGO-CNN-BiGRU-Attention architecture deep learning fusion early warning model in step five specifically includes: searching and locating the globally optimal weights, bias terms, and hyperparameters of the model using the Northern Eagle optimization algorithm; using convolutional neural network layers to perform spatial dimensionality reduction extraction on the multidimensional feature matrix to capture the lateral correlation features between the acoustic and magnetic signals of coal and rock fractures; using bidirectional gated recurrent unit layers to perform forward and backward bidirectional scanning of time-series signals to capture the long short-term memory features of the evolution of mining stress fields; and using attention mechanism layers to dynamically allocate the feature contribution of different sensing modes to achieve a significant enhancement of the precursor signals of coal and rock gas outbursts or rockburst anomalies.

[0024] Furthermore, in step five, the decision-making level fusion of different early warning sources adopts the DS theory based on evidence conflict resolution. In its fusion rules, the evidence body... weight Based on its Bray-Curtis dissimilarity with other evidence Decide: , Where N represents the number of sensor sources;

[0025] Using weights After revising the original evidence, the Dempster combination rule is applied to calculate the final confidence level for each risk level.

[0026] Furthermore, the spatial sampling accuracy of the distributed optical fiber sensor network is set to... to The sampling frequency is set according to the frequency range of coal and rock micro-fractures. to The detection antenna has the function of shielding against interference from the underground power grid, and its effective sensing radius is not less than [amount missing]. .

[0027] Furthermore, during installation, the distributed optical fiber sensing cable is grouted with special cement-based materials or resin anchoring agents for mining, or the cable is incorporated into the anchor bolt / cable to ensure that the interface coupling coefficient between the cable and the coal and rock mass or support structure is better than 0.85.

[0028] Furthermore, the support decision includes outputting rockburst early warning levels, which are divided into four levels: no rockburst hazard, weak rockburst hazard, moderate rockburst hazard, and strong rockburst hazard. For strong rockburst hazard, the system outputs suggestions including: immediately stop mining operations, evacuate personnel, implement large-diameter borehole decompression or coal seam water injection to soften the dangerous area.

[0029] Compared with existing technologies, this invention has the following advantages: First, it achieves complementary synergy in the spatiotemporal dimensions through the full-space coverage of distributed optical fibers and the stress sensitivity of electromagnetic radiation. Optical fiber technology solves the blind spot problem of traditional point monitoring, while the redshift characteristics of the dominant frequency captured by the electromagnetic antenna effectively compensate for the response lag of vibration signals during the "silent period" before rock mass failure, realizing "acoustic-magnetic mutual verification" of disaster precursors. Second, the signal processing layer and the physical sensing layer work closely together, utilizing the entropy difference characteristics of CEEMDAN decomposition and permutation entropy (PE) to accurately identify and remove mechanical construction noise with high entropy characteristics, reconstructing energy without damaging the phase of the microseismic signal. This "physical-level denoising" at the front end provides a high-fidelity data foundation for feature extraction at the back end. More importantly, at the intelligent decision-making layer, the NGO-optimized CNN-BiGRU architecture and the attention mechanism have a deep coupling effect: BiGRU mines long-term and short-term stress evolution memories, while the attention mechanism can dynamically allocate weights based on the real-time contribution of acoustic-magnetic features, ensuring that risks can still be locked when microseismic energy is still low but electromagnetic anomalies are significant, completely solving the problem of missed detection. Finally, the DS evidence theory was introduced to establish a conflict resolution mechanism at the decision-making terminal. When a physical field is subjected to various disturbances and false evidence appears, the system can automatically reduce its fusion weight. This logical self-consistency and fault-tolerant collaboration between levels fundamentally ensures the accuracy of early warning and the robustness of the system in the extremely complex environment of ultra-deep underground engineering. Attached Figure Description

[0030] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the specific embodiments are briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments. It should be noted that the embodiments described herein are merely some embodiments of this invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. Furthermore, in the absence of conflict, the embodiments and features in the embodiments of this invention can be combined with each other.

[0033] Example 1:

[0034] A method for monitoring the support of underground roadways and working faces in coal mines, the method comprising the following steps:

[0035] Step 1: Lay flame-retardant and anti-static distributed optical fiber sensing cables along the roadway axis or in the roadway perimeter of the support structure in the coal mine's longwall mining roadway, tunneling roadway, or fully mechanized mining face. Construct a distributed vibration sensing network based on phase-sensitive optical time-domain reflectometry (PTZ) technology to acquire real-time vibration time-series signals of the entire coal and rock mass and support structure. Simultaneously, install intrinsically safe multi-dimensional electromagnetic radiation detection antennas in the advanced support area and stress concentration area of ​​the mining face to synchronously acquire broadband electromagnetic radiation pulse signals generated by the coal and rock mass under mining stress in real time.

[0036] Step 2: The collected distributed vibration signal is nonlinearly and nonstationarily decomposed using the CEEMDAN algorithm to generate a series of intrinsic mode function components. By calculating the permutation entropy value of each intrinsic mode function component, a component attribute identification matrix is ​​constructed to identify and remove high-entropy mechanical noise components affected by coal mining machine cutting, scraper conveyor operation, and hydraulic support movement, while retaining low-entropy dominant components containing information on coal and rock micro-fracture characteristics. For the retained components, an improved wavelet threshold function is introduced for further noise reduction processing to reconstruct the useful signal of coal and rock micro-vibration.

[0037] Step 3: Extract the spatial location coordinates, energy release rate, dominant frequency, and indicators reflecting the evolution trend of rockbursts from the preprocessed distributed vibration signals. Value sequence; synchronously extract electromagnetic pulse intensity, pulse count rate and main frequency redshift features of electromagnetic waves from electromagnetic radiation signals, and establish a joint feature vector matrix covering acoustic and magnetic features;

[0038] Step 4: Using a pre-defined coupled mathematical model of mining stress-microseismic activity-gas-electromagnetic radiation, input the joint feature vector extracted in Step 3 into this coupled model to invert and obtain the dynamic stress state and damage factor of the coal mine roadway support structure and coal-rock interface. ;

[0039] Step 5: Construct a deep learning fusion early warning model based on the NGO-CNN-BiGRU-Attention architecture. Use multi-physics feature vectors and inverted damage factors as inputs to predict the instability probability and rockburst hazard level of the support structure in the future. Use the DS evidence theory based on Bray-Curtis dissimilarity improvement to fuse the early warning conclusions from fiber optic sensing and electromagnetic sensing at the decision layer, and output the final early warning signal and targeted dynamic support adjustment suggestions.

[0040] Furthermore, in step two, the CEEMDAN algorithm is used to analyze the vibration signal. The specific mathematical implementation process for decomposition is as follows: Define the operator The first generated by empirical mode decomposition Each intrinsic mode component Let i be a standard Gaussian white noise sequence with zero mean and constant variance for the i-th experiment, where i represents the experiment number. The noise coefficient is added when processing the k-th residual signal, and I is the total number of experiments; first, a signal affected by mining noise is constructed. Empirical mode decomposition is performed on it to obtain the first intrinsic mode component. Mean:

[0041]

[0042] Calculate the first residual term Then, an adaptive noise component is added to the residual term, and the 1st term is iteratively calculated. One IMF component:

[0043]

[0044] The complete expansion of the signal is finally obtained until the residual signal no longer satisfies the decomposition conditions. .

[0045] Furthermore, the component selection in step two employs the permutation entropy criterion, and its specific steps include: reconstructing the phase space for each IMF component to obtain a matrix. Its row vectors contain the embedding dimension. and delay time The time series segment; the elements in the row vector are sorted in ascending order, and the probability of each sorting pattern is calculated. The permutation entropy of this component is calculated using the Shannon entropy formula. : Set the permutation entropy threshold It is 0.6, when the component is If the noise is determined to be mechanical vibration noise generated by the coal mining machine or conveyor, it will be removed.

[0046] Furthermore, in step three, when extracting microseismic features, a mechanism reflecting the energy release pattern of coal and rock is introduced. The value calculation formula is based on the statistical law of the magnitude-frequency relationship: ,in, The magnitude is [missing information]. For earthquakes with a magnitude greater than or equal to The cumulative number of microseismic events This is a parameter representing the regional activity level; calculated using a sliding time window. The real-time evolution curve of the value, when When the value shows a significant and continuous decline, it is determined that the working face or roadway is threatened by rock burst.

[0047] Furthermore, in the coupled mathematical model of mining stress-microseismic-electromagnetic radiation in step four, the composite damage factor... The calculation formula is: ,in, This represents the total number of electromagnetic radiation pulses detected per unit time. This represents the maximum instantaneous electromagnetic pulse value at which the coal body reaches its peak strength under the geological conditions of this coal seam.

[0048] Furthermore, the NGO-CNN-BiGRU-Attention architecture deep learning fusion early warning model in step five specifically includes: searching and locating the globally optimal weights, bias terms, and hyperparameters of the model using the Northern Eagle optimization algorithm; using convolutional neural network layers to perform spatial dimensionality reduction extraction on the multidimensional feature matrix to capture the lateral correlation features between the acoustic and magnetic signals of coal and rock fractures; using bidirectional gated recurrent unit layers to perform forward and backward bidirectional scanning of time-series signals to capture the long short-term memory features of the evolution of mining stress fields; and using attention mechanism layers to dynamically allocate the feature contribution of different sensing modes to achieve a significant enhancement of the precursor signals of coal and rock gas outbursts or rockburst anomalies.

[0049] Furthermore, in step five, the decision-making level fusion of different early warning sources adopts the DS theory based on evidence conflict resolution. In its fusion rules, the evidence body... weight Based on its Bray-Curtis dissimilarity with other evidence Decide: , Where N represents the number of sensor sources;

[0050] Using weights After revising the original evidence, the Dempster combination rule is applied to calculate the final confidence level for each risk level.

[0051] Furthermore, the spatial sampling accuracy of the distributed optical fiber sensor network is set to... to The sampling frequency is set according to the frequency range of coal and rock micro-fractures. to The detection antenna has the function of shielding against interference from the underground power grid, and its effective sensing radius is not less than [amount missing]. .

[0052] Furthermore, during installation, the distributed optical fiber sensing cable is grouted with special cement-based materials or resin anchoring agents for mining, or the cable is incorporated into the anchor bolt / cable to ensure that the interface coupling coefficient between the cable and the coal and rock mass or support structure is better than 0.85.

[0053] Furthermore, the support decision includes outputting rockburst early warning levels, which are divided into four levels: no rockburst hazard, weak rockburst hazard, moderate rockburst hazard, and strong rockburst hazard. For strong rockburst hazard, the system outputs suggestions including: immediately stop mining operations, evacuate personnel, implement large-diameter borehole decompression or coal seam water injection to soften the dangerous area.

[0054] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring the support of underground roadways and working faces in coal mines, characterized in that, The method includes the following steps: Step 1: Lay flame-retardant and anti-static distributed optical fiber sensing cables along the roadway axis or in the roadway perimeter of the support structure in the coal mine's longwall roadway, tunneling roadway, or fully mechanized mining face to construct a distributed vibration sensing network based on phase-sensitive optical time-domain reflectometry technology, which is used to acquire the full-line vibration time sequence signal of the coal and rock mass and the support structure in real time; At the same time, install intrinsically safe multi-dimensional electromagnetic radiation detection antennas in the advanced support area and stress concentration area of ​​the mining face to synchronously acquire the broadband electromagnetic radiation pulse signal generated by the coal and rock mass under mining stress in real time. Step 2: The collected distributed vibration signal is nonlinearly and nonstationarily decomposed using the CEEMDAN algorithm to generate a series of intrinsic mode function components. By calculating the permutation entropy value of each intrinsic mode function component, a component attribute identification matrix is ​​constructed to identify and remove high-entropy mechanical noise components affected by coal mining machine cutting, scraper conveyor operation, and hydraulic support movement, while retaining low-entropy dominant components containing information on coal and rock micro-fracture characteristics. For the retained components, an improved wavelet threshold function is introduced for further noise reduction processing to reconstruct the useful signal of coal and rock micro-vibration. Step 3: Extract the spatial location coordinates, energy release rate, dominant frequency, and indicators reflecting the evolution trend of rockbursts from the preprocessed distributed vibration signals. Value sequence; synchronously extract electromagnetic pulse intensity, pulse count rate and main frequency redshift features of electromagnetic waves from electromagnetic radiation signals, and establish a joint feature vector matrix covering acoustic and magnetic features; Step 4: Using a pre-defined coupled mathematical model of mining stress-microseismic activity-gas-electromagnetic radiation, input the joint feature vector extracted in Step 3 into this coupled model to invert and obtain the dynamic stress state and damage factor of the coal mine roadway support structure and coal-rock interface. ; Step 5: Construct a deep learning fusion early warning model based on the NGO-CNN-BiGRU-Attention architecture. Use multi-physics feature vectors and inverted damage factors as inputs to predict the instability probability and rockburst hazard level of the support structure in the future. Use the DS evidence theory based on Bray-Curtis dissimilarity improvement to fuse the early warning conclusions from fiber optic sensing and electromagnetic sensing at the decision layer, and output the final early warning signal and targeted dynamic support adjustment suggestions.

2. The method according to claim 1, characterized in that, In step two, the CEEMDAN algorithm is used to analyze the vibration signal. The specific mathematical implementation process for decomposition is as follows: Define the operator The first generated by empirical mode decomposition Each intrinsic mode component Let i be a standard Gaussian white noise sequence with zero mean and constant variance for the i-th experiment, where i represents the experiment number. The noise coefficient is added when processing the k-th residual signal, and I is the total number of experiments; first, a signal affected by mining noise is constructed. Empirical mode decomposition is performed on it to obtain the first intrinsic mode component. Mean: ， Calculate the first residual term Then, an adaptive noise component is added to the residual term, and the 1st term is iteratively calculated. One IMF component: ， The complete expansion of the signal is finally obtained until the residual signal no longer satisfies the decomposition conditions. , where K represents the final number of layers the signal is decomposed into.

3. The method according to claim 2, characterized in that, The selection of components in step two employs the permutation entropy criterion, and its specific steps include: For each IMF component, phase space reconstruction is performed to obtain the matrix. Its row vectors contain the embedding dimension. and delay time The time series segment; the elements in the row vector are sorted in ascending order, and the probability of each sorting pattern is calculated. The permutation entropy of this component is calculated using the Shannon entropy formula. : Set the permutation entropy threshold It is 0.6, when the component is If the noise is determined to be mechanical vibration noise generated by the coal mining machine or conveyor, it will be removed.

4. The method according to claim 1, characterized in that, When extracting microseismic features in step three, a mechanism reflecting the energy release pattern of coal and rock is introduced. The value calculation formula is based on the statistical law of the magnitude-frequency relationship: ,in, The magnitude is [missing information]. For earthquakes with a magnitude greater than or equal to The cumulative number of microseismic events This is a parameter representing the regional activity level; calculated using a sliding time window. The real-time evolution curve of the value, when When the value shows a significant and continuous decline, it is determined that the working face or roadway is threatened by rock burst.

5. The method according to claim 1, characterized in that, In the mining stress-microseismic-electromagnetic radiation coupled mathematical model in step four, the composite damage factor The calculation formula is: ,in, This represents the total number of electromagnetic radiation pulses detected per unit time. This represents the maximum instantaneous electromagnetic pulse value at which the coal body reaches its peak strength under the geological conditions of this coal seam.

6. The method according to claim 1, characterized in that, The NGO-CNN-BiGRU-Attention architecture deep learning fusion early warning model in step five specifically includes: searching and locating the model's globally optimal weights, bias terms, and hyperparameters using the Northern Eagle optimization algorithm; extracting the multidimensional feature matrix using a convolutional neural network layer to capture the lateral correlation features between the acoustic and magnetic signals of coal and rock fractures; performing forward and backward bidirectional scanning of the time-series signal using a bidirectional gated recurrent unit layer to capture the long short-term memory features of the mining stress field evolution; and dynamically allocating the weights of the feature contributions of different sensing modes through an attention mechanism layer to significantly enhance the precursor signals of coal and rock gas outbursts or rockburst anomalies.

7. The method according to claim 1, characterized in that, In step five, the decision-making level fusion of different early warning sources adopts the DS theory based on evidence conflict resolution. In its fusion rules, the evidence body... weight Based on its Bray-Curtis dissimilarity with other evidence Decide: , Where N represents the number of sensor sources; using weights After revising the original evidence, the Dempster combination rule is applied to calculate the final confidence level for each risk level.

8. The method according to claim 1, characterized in that, The spatial sampling accuracy of the distributed optical fiber sensing network is set to... to The sampling frequency is set according to the frequency range of coal and rock micro-fractures. to The detection antenna has the function of shielding against interference from the underground power grid, and its effective sensing radius is not less than [amount missing]. .

9. The method according to claim 1, characterized in that, During installation, the distributed optical fiber sensing cable is grouted with special cement-based materials or resin anchoring agents for mining, or the cable is incorporated into the anchor bolt / cable to ensure that the interface coupling coefficient between the cable and the coal and rock mass or support structure is better than 0.

85.

10. The method according to claim 1, characterized in that, The support decision includes outputting rockburst early warning levels, which are divided into four levels: no rockburst risk, weak rockburst risk, moderate rockburst risk, and strong rockburst risk. In response to the risk of severe impact, the system's recommendations include: immediately halting mining operations, evacuating personnel, and implementing pressure relief through large-diameter boreholes or water injection to soften the coal seam in the hazardous area.

Images