Coal mine underground and roadway three-dimensional earthquake dynamic monitoring system based on AI

Abstract

The invention relates to the technical field of dynamic monitoring, in particular to an AI-based coal mine underground and roadway three-dimensional earthquake dynamic monitoring system, which comprises an underground edge computing node cluster module, an underground edge computing node cluster module and an underground edge computing node cluster module, the multi-source heterogeneous data fusion processing module is used for receiving the original signal data set, performing space-time alignment and feature decoupling processing, and generating a standardized seismic-electric feature matrix; the three-dimensional geological dynamic modeling module is used for generating a three-dimensional geological dynamic model updated in real time through a dynamic voxel compensation algorithm based on the standardized seismoelectric characteristic matrix; the abnormal characteristic evolution analysis module is used for receiving time sequence data of the three-dimensional geological dynamic model and constructing a space-time two-dimensional abnormal degree evolution graph; and the visual early warning decision module is used for generating graded early warning signals and a three-dimensional visual interface. According to the invention, the six-level progressive processing architecture is innovatively designed, so that the performance of the three-dimensional earthquake dynamic monitoring system for the underground coal mine and the roadway is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic monitoring, and particularly to a three-dimensional seismic dynamic monitoring system for coal mines underground and roadways based on AI. Background Art

[0002] The geological environment underground in coal mines is complex, and disasters such as earthquakes and rock bursts occur frequently, posing a serious threat to safe production. Traditional monitoring systems mostly rely on single sensors for static monitoring, with significant technical bottlenecks. First, the existing systems do not fully consider the electro-seismic coupling effect, resulting in difficult effective stripping of electromechanical interference signals and serious distortion of signal characteristics. Second, traditional three-dimensional geological modeling relies on regularly surveyed artificial data, with an update cycle of up to several hours, and cannot reflect geological dynamic changes in real time. These problems severely restrict the accuracy and timeliness of coal mine disaster early warning.

[0003] Existing technologies have tried to introduce AI algorithms to improve monitoring performance, but there are still significant defects. Traditional methods are based on static threshold judgment and do not consider the evolution law of geological anomalies in the spatio-temporal dimension. In addition, existing systems mostly use two-dimensional charts to display monitoring results, which cannot intuitively reflect the spatial distribution and evolution trend of anomalies, and it is difficult for underground personnel to quickly understand the risk situation. At the same time, there is a lack of a compensation mechanism for blind area data, resulting in insufficient monitoring coverage in key areas. These limitations make it difficult for existing systems to meet the high-precision and real-time monitoring requirements in deep mining areas. Summary of the Invention

[0004] Based on the above purpose, the present invention provides a three-dimensional seismic dynamic monitoring system for coal mines underground and roadways based on AI.

[0005] The three-dimensional seismic dynamic monitoring system for coal mines underground and roadways based on AI includes the following modules:

[0006] Underground edge computing node cluster module: Edge computing nodes are arranged in the mining face and the transportation roadway to synchronously collect seismic wave signals, electromagnetic radiation signals, and microseismic signals.

[0007] Multi-source heterogeneous data fusion processing module: Receives the original signal data set output by the underground edge computing node cluster module, performs spatio-temporal alignment and feature decoupling processing, and generates a standardized electro-seismic feature matrix.

[0008] Three-dimensional geological dynamic modeling module: Based on the standardized electro-seismic feature matrix, combined with the three-dimensional point cloud data of the roadway, generates a real-time updated three-dimensional geological dynamic model through a dynamic voxel compensation algorithm.

[0009] Abnormal feature evolution analysis module: Receives the time series data of the three-dimensional geological dynamic model and constructs a spatio-temporal two-dimensional anomaly degree evolution map.

[0010] Visual warning decision-making module: Match the spatio-temporal two-dimensional anomaly evolution map with a preset geological structure database to generate hierarchical warning signals and a three-dimensional visualization interface.

[0011] Furthermore, the underground edge computing node cluster module includes:

[0012] Topological optimization of node layout: Layout the mining face in a diamond grid;

[0013] Synchronous acquisition of multi-source signals: Achieve time alignment of heterogeneous sensors through a clock drift compensation algorithm, and calculate the corrected acquisition time;

[0014] Signal coupling suppression: Perform mechanical and electrical interference suppression processing on the signals collected by electromagnetic radiation sensors;

[0015] Dynamic sampling rate adjustment: Adaptively adjust the sampling rate of microseismic sensors according to the signal frequency band characteristics.

[0016] Furthermore, the multi-source heterogeneous data fusion processing module includes:

[0017] Spatio-temporal alignment of multi-source signals: Perform spatio-temporal registration on the original signal data set to obtain signals after spatio-temporal alignment;

[0018] Decoupling of seismic-electric coupling characteristics: Jointly analyze the signals after spatio-temporal alignment, construct a seismic-electric coupling tensor, and achieve feature decoupling through high-order singular value decomposition;

[0019] Generation of a standardized seismic-electric feature matrix: Normalize the decoupled eigenvectors to generate a standardized seismic-electric feature matrix.

[0020] Furthermore, the generation of the standardized seismic-electric feature matrix includes:

[0021] Extraction of eigenvectors: Extract decoupled eigenvectors from the results of high-order singular value decomposition;

[0022] Multi-modal fusion and standardization: Perform multi-modal fusion and standardization on the eigenvectors.

[0023] Furthermore, the three-dimensional geological dynamic modeling module includes:

[0024] Initialization of the voxel space: Convert the three-dimensional point cloud data of the roadway into an initial voxel grid, where the density value of the voxel is determined by the spatial distribution of the point cloud;

[0025] Seismic-electric feature-voxel mapping: Based on the standardized seismic-electric feature matrix, establish a projection relationship between the sensor nodes and the voxel space, and calculate the feature enhancement value of the voxel;

[0026] Blind area voxel compensation: For the voxel area not covered by the sensor, a spatio-temporal LSTM network is used to predict the compensation value and calculate the compensated voxels.

[0027] Dynamic model update: Integrate physical measurements and predicted compensation results to generate a real-time three-dimensional geological model.

[0028] Furthermore, the abnormal feature evolution analysis module includes:

[0029] Spatio-temporal feature field construction and standardization preprocessing: Based on the sequence of three-dimensional geological dynamic models within continuous time periods, construct a four-dimensional feature field that fuses the spatial dimension and the time dimension, and perform dynamic baseline calibration in the time dimension for each spatial unit.

[0030] Spatio-temporal two-dimensional anomaly detection: Use the adaptive time window decomposition technique to separate the sudden mutations and gradual changes of geological parameters, quantify the anomaly intensity in the time dimension through residual ratio analysis, and design a three-dimensional convolutional kernel driven by geomechanics to detect the spatial anomaly intensity.

[0031] Abnormal evolution map generation and optimization: Dynamically weight and fuse the time anomaly intensity and the spatial anomaly intensity to generate a spatio-temporal two-dimensional anomaly degree map.

[0032] Furthermore, the spatio-temporal feature field construction and standardization preprocessing convert the sequence of three-dimensional geological dynamic models at consecutive moments into four-dimensional feature vectors, and perform standardization in the time dimension for each voxel.

[0033] Furthermore, the spatio-temporal two-dimensional anomaly detection includes:

[0034] Time dimension anomaly degree calculation: Use the adaptive sliding time window decomposition to extract the trend term, periodic term, and residual term, and calculate the time anomaly index.

[0035] Spatial dimension anomaly degree calculation: Construct an anisotropic three-dimensional convolutional kernel;

[0036] Calculate the spatial anomaly degree through convolution.

[0037] Furthermore, the abnormal evolution map generation and optimization includes:

[0038] Spatio-temporal anomaly degree fusion: Dynamically weight and fuse the time and spatial anomaly intensities to generate a spatio-temporal two-dimensional anomaly degree map;

[0039] Abnormal evolution map generation: Perform non-maximum suppression and morphological optimization on the spatio-temporal two-dimensional anomaly degree map, extract the connected abnormal regions and encode the evolution paths.

[0040] Furthermore, the visualization warning decision module includes:

[0041] Geological anomaly pattern matching: Construct a similarity measurement model enhanced by dynamic time warping to calculate the matching degree between the real-time anomaly map and the geological structure database;

[0042] Hierarchical early warning signal generation: Generate three-level early warnings based on the matching degree and the anomaly intensity gradient;

[0043] Three-dimensional enhanced visualization: Use mixed reality technology to generate a dynamic early warning interface.

[0044] Advantages of the present invention:

[0045] Through an innovative six-layer progressive processing architecture, the present invention significantly improves the performance of the three-dimensional seismic dynamic monitoring system for coal mines underground and in roadways. First, the system uses an underground edge computing node cluster and multi-source heterogeneous data fusion technology to achieve synchronous acquisition and precise decoupling of seismic and electrical signals. Second, through the dynamic voxel compensation algorithm and the spatio-temporal two-dimensional anomaly evolution model, the system effectively solves the problems of blind area modeling in roadways and prediction of anomaly propagation paths, reducing the update delay of the three-dimensional model and improving the anomaly positioning accuracy.

[0046] By constructing a geological structure knowledge graph and augmented reality projection technology, the system can spatially superimpose and display the early warning area and the actual roadway video stream, supporting a centimeter-level spatial registration accuracy. At the same time, the self-calibration and optimization module dynamically adjusts the algorithm parameters according to the deviation data between historical early warning records and real geological structures, reducing the false alarm rate of the system monthly. These innovations not only improve the cognitive efficiency of underground personnel towards the risk situation but also provide intelligent decision-making support for coal mine safety production, with significant economic and social benefits. Description of the Drawings

[0047] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 It is the system module diagram of the embodiment of the present invention;

[0049] Figure 2 It is the node calculation diagram of the embodiment of the present invention. Detailed Embodiments

[0050] To make the purpose, technical solutions, and advantages of the present invention clearer, the following further details the present invention with specific embodiments.

[0051] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this term cover the elements or objects listed after this term and their equivalents, without excluding other elements or objects. The terms such as "connected" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left" and "right" are only used to represent relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0052] As Figure 1 - Figure 2 shown, the AI-based three-dimensional seismic dynamic monitoring system for coal mines underground and in roadways includes the following modules:

[0053] Underground edge computing node cluster module: Edge computing nodes are deployed in the mining face and transportation roadway to synchronously collect seismic wave signals, electromagnetic radiation signals, and microseismic signals;

[0054] Multi-source heterogeneous data fusion processing module: Receives the original signal data set output by the underground edge computing node cluster module, performs spatio-temporal alignment and feature decoupling processing, and generates a standardized seismic-electric feature matrix;

[0055] Three-dimensional geological dynamic modeling module: Based on the standardized seismic-electric feature matrix, combined with the three-dimensional point cloud data of the roadway, generates a real-time updated three-dimensional geological dynamic model through a dynamic voxel compensation algorithm;

[0056] Abnormal feature evolution analysis module: Receives the time series data of the three-dimensional geological dynamic model and constructs a spatio-temporal two-dimensional abnormal degree evolution map;

[0057] Visualization warning decision module: Matches the spatio-temporal two-dimensional abnormal degree evolution map with a preset geological structure database to generate a hierarchical warning signal and a three-dimensional visualization interface.

[0058] The underground edge computing node cluster module includes:

[0059] Node layout topology optimization: The mining face is arranged in a diamond grid, and the adjacent node spacing is expressed as:

[0060]

[0061] where D is the node spacing (m), H is the working face height (m), v s is the shear wave velocity of the coal seam (m / s), fmax is the highest effective frequency (Hz) of the seismic signal;

[0062] Nodes are arranged at intervals of L meters along the roof center line in the transportation roadway. L is jointly determined by the radius of curvature R (m) of the roadway and the seismic wave wavelength λ (m), and is expressed as:

[0063]

[0064] Multi-source signal synchronous acquisition: Time alignment of heterogeneous sensors is achieved through the clock drift compensation algorithm, and the corrected acquisition time t i ′ is calculated and expressed as:

[0065] t i ′ = α i t i + β i +(t);

[0066] Among them, Δt i is the local clock deviation of the i-th node, t ref is the global reference clock, α i is the clock rate correction factor (dimensionless), β i is the clock phase offset, and (t) is the random error term (s) introduced by the environmental temperature drift, which is estimated and eliminated in real time through Kalman filtering;

[0067] Signal coupling suppression: The signal collected by the electromagnetic radiation sensor is processed to suppress electromechanical interference, and is expressed as:

[0068]

[0069] Among them, A k , f k , φ k are respectively the amplitude (V / m), characteristic frequency (Hz), and phase angle (rad) of the k-th type of electromechanical interference, which are extracted through real-time spectrum analysis, and E(t) is the signal collected by the magnetic radiation sensor;

[0070] Dynamic sampling rate adjustment: The sampling rate F of the microseismic sensor is adaptively adjusted according to the signal frequency band characteristics s and is expressed as:

[0071]

[0072] Among them, σ is the root mean square value (Pa) of the microseismic signal, and σ max is the historical maximum root mean square value (Pa).

[0073] The multi-source heterogeneous data fusion processing module includes:

[0074] Multi-source signal spatio-temporal alignment: Spatio-temporal registration is performed on the original signal dataset to obtain the spatio-temporally aligned signal, expressed as:

[0075]

[0076] where S i (t) is the seismic wave signal of the i-th node, the electromagnetic radiation signal is E i (t), the microseismic signal is M i (t), w ik is the spatial weight coefficient, determined by the spatial distance d ik (m) between node i and adjacent node k, σ is the sensor spatial influence radius (m), taking a value of 1.5 times the average node spacing, Δt ik is the time compensation amount (s), jointly determined by the signal propagation time delay and the clock deviation, v p is the longitudinal wave velocity (m / s) of the seismic wave in the coal seam, δt ik is the clock deviation (s) between nodes;

[0077] Seismo-electric coupling feature decoupling: Joint analysis is performed on the spatio-temporally aligned signal to construct the seismo-electric coupling tensor, and feature decoupling is achieved through high-order singular value decomposition (HOSVD), expressed as:

[0078] T = G × 1U (1) × 2U (2) × 3U (3) ;

[0079] where, is the seismo-pad coupling tensor, N is the number of nodes, T is the time series length, the 3 channels respectively correspond to the seismic wave, electromagnetic radiation, and microseismic signals, G is the core tensor, are the spatial, time, and modal factor matrices respectively, retaining the first R1 = 5, R2 = 10, R3 = 2 principal components to extract the coupling features;

[0080] Standardized seismo-electric feature matrix generation: The decoupled eigenvectors are normalized to generate the standardized seismo-electric feature matrix.

[0081] Standardized seismo-electric feature matrix generation includes:

[0082] Eigenvector extraction: Decoupled eigenvectors are extracted from the high-order singular value decomposition results, expressed as:

[0083] V q = vec(G(q, :, :) × 2U (2) × 3U (3) )(q = 1,..., N);

[0084] Among them, G(q, :, :) represents the q-th slice of the core tensor in the first dimension (spatial dimension), and vec(·) represents the tensor vectorization operation to obtain the eigenvector corresponding to each node.

[0085] Multimodal fusion and normalization: The eigenvectors are subjected to multimodal fusion and normalization, expressed as:

[0086]

[0087] Among them, is the eigenvalue of the i-th node in the j-th feature dimension and the k-th modality (seismic / electromagnetic / microseismic), ω k is the modality weight coefficient (ω1 = 0.6, ω2 = 0.3, ω3 = 0.1), μ j , σ j are the feature mean and standard deviation calculated based on historical data, and η = 0.15 is the electromagnetic compensation factor used to enhance the sensitivity to mutation signals.

[0088] The three-dimensional geological dynamic modeling module includes:

[0089] Voxel space initialization: Convert the three-dimensional point cloud data of the roadway (N p is the number of point clouds) into an initial voxel grid where the density value ρ lwh of the voxel v lwh is determined by the spatial distribution of the point cloud, expressed as:

[0090]

[0091] Among them, c lwh is the voxel center coordinate (m), σ p = 0.3D v is the smoothing coefficient (D v is the voxel side length);

[0092] Seismo-electric feature - voxel mapping: Based on the standardized seismo-electric feature matrix Establish the projection relationship between the sensor nodes and the voxel space, and calculate the feature enhancement value Δρ lwh of the voxel v lwh , expressed as:

[0093]

[0094] Among them, Ω i is the voxel region corresponding to the i-th node, is the set of neighboring nodes of the voxel v lwh , is the trainable weight matrix, s iis the spatial coordinate of node i (m), v s is the shear wave velocity (m / s), τ = 0.2 s is the signal attenuation time constant;

[0095] Blind area voxel compensation: For the voxel region Ω not covered by the sensor blind , use the spatio-temporal LSTM network to predict the compensation value and calculate the compensated voxel It is expressed as:

[0096]

[0097] where t is the current time, is the historical voxel sequence, LSTM θ is the pre-trained transfer learning network, M geo ∈[0,1] L×W×H is the geological similarity mask, constructed from the historical data of adjacent roadways;

[0098] Dynamic model update: Integrate physical measurements and prediction compensation results to generate a real-time three-dimensional geological model V t , which is expressed as:

[0099]

[0100] where α = 0.7 is the confidence level of measured data, β = 0.05 is the time gradient weight coefficient, is the time derivative of the seismo-electric characteristic.

[0101] The abnormal feature evolution analysis module includes:

[0102] Spatio-temporal feature field construction and standardization preprocessing: Based on the sequence of three-dimensional geological dynamic models within a continuous time period, construct a four-dimensional feature field that integrates the spatial dimension and the time dimension, and perform dynamic baseline calibration in the time dimension for each spatial unit;

[0103] Spatio-temporal two-dimensional anomaly detection: Use the adaptive time window decomposition technology to separate the sudden mutations and progressive changes of geological parameters, quantify the anomaly intensity in the time dimension through residual ratio analysis, and design a three-dimensional convolutional kernel driven by geomechanics to detect the spatial anomaly intensity;

[0104] Abnormal evolution map generation and optimization: Dynamically weight and fuse the time anomaly intensity and the spatial anomaly intensity to generate a spatio-temporal two-dimensional anomaly degree map.

[0105] The spatio-temporal feature field construction and standardization preprocessing converts the sequence of three-dimensional geological dynamic models at consecutive moments into four-dimensional feature vectors and performs standardization in the time dimension for each voxel, which is expressed as:

[0106]

[0107] Among them, v lwh is a voxel, is a sequence of three-dimensional geological dynamic models at T consecutive time moments.

[0108] The spatio-temporal two-dimensional anomaly detection includes:

[0109] Calculation of anomaly degree in the time dimension: The trend term periodic term and residual term are extracted by using adaptive sliding time window decomposition, and the time anomaly index is calculated, expressed as:

[0110]

[0111] Among them, τ = 10 is the time window length (seconds), ∈ = 1e -5 is a smoothing term to prevent division by zero;

[0112] Calculation of anomaly degree in the space dimension: An anisotropic three-dimensional convolution kernel with a weight distribution conforming to the coal seam stress propagation characteristics is constructed, and the three-dimensional convolution kernel is expressed as:

[0113]

[0114] Among them, Δx = Δy = 0.5m, Δz = 0.3m is the voxel spacing, θ = 15° is the coal seam dip angle, σ g = 1.2m is the geological continuity factor, and Z is the normalization factor;

[0115] The space anomaly degree is calculated through convolution, expressed as:

[0116]

[0117] Among them, represents the space gradient operator.

[0118] The generation and optimization of the anomaly evolution map include:

[0119] Spatio-temporal anomaly degree fusion: Dynamically weighted fusion of the time and space anomaly intensities is performed to generate a spatio-temporal two-dimensional anomaly degree map expressed as:

[0120]

[0121] Among them, the dynamic weight coefficient λ(t) is determined by the historical anomaly pattern matching degree, σ(·) is the sigmoid function, w = [0.5, 0.3, 0.2] is the time decay weight, sim(·) is the distance similarity, and γ = 0.1 is the time gradient gain;

[0122] Abnormal evolution graph generation: Perform non-maximum suppression and morphological optimization on the spatio-temporal two-dimensional anomaly degree graph A, extract connected abnormal regions and encode the evolution paths, expressed as:

[0123]

[0124] Among them, Geodesic(·) is the geodesic propagation path calculated based on the three-dimensional roadway surface, G(t) is the abnormal evolution graph at time t, and K is the total number of connected abnormal regions.

[0125] The visualization and early warning decision-making module includes:

[0126] Geological anomaly pattern matching: Construct a similarity measurement model enhanced by dynamic time warping, and calculate the real-time anomaly graph A t and the geological structure database matching degree, expressed as:

[0127]

[0128] Among them, is the normalization factor, δ = 1e -3 is the smoothing coefficient, is the spatial weight, s key is the central coordinate (m) of the key monitoring area, T resp = 2s is the system response time threshold;

[0129] Three-level early warning signal generation: Generate three-level early warnings based on the matching degree and the abnormal intensity gradient, expressed as:

[0130]

[0131] Among them, A max = 50 is the maximum anomaly degree, calculated by the five-point difference method;

[0132] Three-dimensional enhanced visualization: Use mixed reality technology to generate a dynamic early warning interface, expressed as:

[0133]

[0134] Among them, α = [0.4, 0.3, 0.3] is the transparency coefficient of the roadway, rock formation, and equipment, β = 0.7 is the heat map rendering intensity, γ = 0.2 represents the Laplacian operator boundary enhancement coefficient, is the geological model, β·Colormap(A t (x, y, z)) represents the abnormal heat map, represents boundary strengthening.

[0135] Those of ordinary skill in the art should understand that: The discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the present invention is limited to these examples; Under the concept of the present invention, the technical features in the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations in different aspects of the present invention as described above, which are not provided in detail for the sake of brevity.

Claims

1. AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels, characterized by: Includes the following modules: Underground edge computing node cluster module: edge computing nodes are deployed in mining working faces and transportation tunnels to synchronously collect seismic wave signals, electromagnetic radiation signals, and microseismic signals; Multi-source heterogeneous data fusion processing module: receives the original signal data set output by the downhole edge computing node cluster module, performs time-space alignment and feature decoupling processing, and generates a standardized seismic-electric feature matrix; 3D geological dynamic modeling module: Based on the standardized seismic and electrical characteristic matrix and combined with the 3D point cloud data of the tunnel, a real-time updated 3D geological dynamic model is generated through a dynamic voxel compensation algorithm; Abnormal feature evolution analysis module: receives time series data of three-dimensional geological dynamic model and constructs a two-dimensional spatial and temporal abnormal degree evolution map; Visualized early warning decision module: the time-space dual-dimensional anomaly evolution map is matched with the preset geological structure database to generate graded early warning signals and a three-dimensional visualization interface.

2. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 1 is characterized in that: The downhole edge computing node cluster module includes: Node layout topology optimization: layout of mining working face with diamond grid; Synchronous acquisition of multi-source signals: Time alignment of heterogeneous sensors is achieved through clock drift compensation algorithm, and the corrected acquisition time is calculated; Signal coupling suppression: Electromechanical interference suppression processing is performed on the signals collected by the electromagnetic radiation sensor; Dynamic sampling rate adjustment: Adaptively adjust the microseismic sensor sampling rate according to the signal frequency band characteristics.

3. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 2 is characterized in that: The multi-source heterogeneous data fusion processing module includes: Multi-source signal spatiotemporal alignment: perform spatiotemporal registration on the original signal data set to obtain the spatiotemporal aligned signal; Seismoelectric coupling feature decoupling: Jointly analyze the time-space aligned signals, construct the seismoelectric coupling tensor, and achieve feature decoupling through high-order singular value decomposition; Generation of standardized seismoelectric characteristic matrix: Normalize the decoupled eigenvectors to generate a standardized seismoelectric characteristic matrix.

4. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 3 is characterized in that: The generation of the standardized seismoelectric characteristic matrix includes: Feature vector extraction: extract decoupled feature vectors from high-order singular value decomposition results; Multimodal fusion and standardization: Feature vectors are subjected to multimodal fusion and standardization.

5. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 4 is characterized in that: The three-dimensional geological dynamic modeling module includes: Voxel space initialization: converting the roadway 3D point cloud data into an initial voxel grid, where the density of the voxels is determined by the point cloud spatial distribution; Seismoelectric feature-voxel mapping: Based on the standardized seismoelectric feature matrix, the projection relationship between the sensor node and the voxel space is established, and the feature enhancement value of the voxel is calculated; Blind area voxel compensation: For the voxel area not covered by the sensor, the spatiotemporal LSTM network is used to predict the compensation value and calculate the compensation voxel; Dynamic model update: Integrate physical measurement and prediction compensation results to generate real-time 3D geological models.

6. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 5 is characterized in that: The abnormal feature evolution analysis module includes: Construction and standardization preprocessing of spatiotemporal characteristic fields: Based on the three-dimensional geological dynamic model sequence in a continuous time period, a four-dimensional characteristic field integrating the spatial dimension and the temporal dimension is constructed, and a dynamic baseline calibration of the temporal dimension is performed for each spatial unit; Anomaly detection in both time and space: Adaptive time window decomposition technology is used to separate sudden changes and gradual changes in geological parameters, the intensity of anomalies in the time dimension is quantified through residual ratio analysis, and a geomechanics-driven three-dimensional convolution kernel is designed to detect the intensity of spatial anomalies; Anomaly evolution map generation and optimization: Dynamically weighted fusion of temporal anomaly intensity and spatial anomaly intensity to generate a dual-dimensional spatiotemporal anomaly map.

7. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 6 is characterized in that: The spatiotemporal feature field construction and standardization preprocessing converts the three-dimensional geological dynamic model sequence at continuous moments into a four-dimensional feature vector, and standardizes the time dimension of each voxel.

8. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 6 is characterized in that: The spatiotemporal dual-dimensional anomaly detection includes: Calculation of abnormality in time dimension: Adaptive sliding window decomposition is used to extract trend items, cycle items and residual items, and calculate the time anomaly index; Calculation of spatial dimension anomaly: construct anisotropic three-dimensional convolution kernel; Compute spatial outlier through convolution.

9. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 6 is characterized in that: The abnormal evolution map generation and optimization includes: Space-time anomaly fusion: Dynamically weighted fusion of time and space anomaly intensities to generate a space-time dual-dimensional anomaly map; Anomaly evolution map generation: Non-maximum suppression and morphological optimization are performed on the spatiotemporal dual-dimensional anomaly map to extract connected anomaly regions and encode the evolution path.

10. The AI-based three-dimensional seismic dynamic monitoring system for underground coal mines and tunnels according to claim 9 is characterized in that: The visual early warning decision module includes: Geological anomaly pattern matching: construct a similarity measurement model enhanced by dynamic time regularization to calculate the matching degree between the real-time anomaly map and the geological structure database; Generation of graded warning signals: Generate three-level warnings based on matching degree and abnormal intensity gradient; Three-dimensional enhanced visualization: Use mixed reality technology to generate a dynamic warning interface.

Images