Hidden structure detection system for intelligent monitoring, early warning and prevention of mine water disasters

Abstract

The invention relates to the technical field of coal mine water disaster prevention and control, and particularly discloses a hidden structure detection system for mine water disaster intelligent monitoring, early warning and prevention, which is used for solving the problems of low mine hidden structure detection precision, mine water disaster early warning lag and large seismic wave propagation path simulation error. Comprising a multi-source data comprehensive detection module, a real-time imaging module, a micro-seismic multi-scale intelligent analysis module, a seismic wave signal identification module, a regional difference analysis module and an intelligent data processing module. According to the invention, through fusion of multi-source data, real-time imaging, micro-seismic analysis, seismic wave identification and intelligent data processing technologies, precise detection of a mine hidden structure, high-precision prediction of mine earthquakes and water disasters and real-time risk early warning are realized, and timeliness and accuracy of mine water disaster prevention and control are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine water disaster prevention and control. More specifically, the present invention relates to a concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters. Background Art

[0002] Coal resources are one of the main energy sources in China. The resources are rich and widely distributed geographically. Moreover, the hydrogeological conditions vary greatly in each coalfield area, and the mine water disasters show obvious diversification and complexity characteristics. According to statistics, China is one of the countries with the most serious mine water disasters in the world. Therefore, the prevention and control of coal mine water disasters is an important task that needs to be emphasized and solved at present. Concealed structures are geological structures or geological phenomena that are not easily detectable underground or below the surface. Usually, it is difficult to directly observe or detect them, but they play a crucial decisive role in the formation of mine water inrush channels. Factors such as too deep strata, surface coverings, and geological diversity will all lead to the difficulty of predicting concealed structures. Therefore, the intelligent detection of concealed structures has become a huge challenge in the current process of coal mine water disaster prevention and control. The Chinese invention patent with the application number 2020106010182 discloses a device and method for omnidirectional advanced detection of concealed water disasters at the bottom of a borehole. By using a time-domain electromagnetic detection device with multi-source variable current combined emission and multi-component parallel reception, it is ensured that the device can observe the low-resistance abnormal body in front of the device omnidirectionally without moving, and finally realizes the advanced geological prediction of concealed water disasters in front of the bottom of the borehole, solving the problems of insufficient detection borehole space in coal mines, the detection equipment cannot move or rotate in the plane, and cannot perform two-dimensional or three-dimensional observations to obtain sufficient background data, laying a foundation for the detailed distribution of geological anomalies in front. However, with the development and progress of detection technologies, more and more advanced intelligent detection means have been introduced to help study and understand concealed structures more accurately. Among them, the identification of seismic wave signals and the reasons for regional differences involve multiple factors such as underground geological structures, seismic wave propagation, signal processing, and data analysis, which will lead to signal differences. The underground environment will interfere with the signals, and different underground structures cannot be distinguished by current equipment, and it is impossible to monitor and early warn the mine water disasters caused by concealed structures in time, so as to take prevention and control measures in time to reduce casualties and engineering losses. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, the present invention provides a concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters, by

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters, including a multi-source data comprehensive detection module, a real-time imaging module, a microseismic multi-scale intelligent analysis module, a seismic wave signal recognition module, a regional difference analysis module and an intelligent data processing module. The multi-source data comprehensive detection module uses differential GPS positioning technology, integrates Kalman filtering and adaptive noise reduction algorithms, and constructs a three-dimensional geological structure model by using Kriging interpolation and joint inversion methods. The real-time imaging module constructs a real-time underground structure imaging by using an adaptive non-linear regularization inversion method, and uses GPU parallel computing to accelerate the imaging process. The microseismic multi-scale intelligent analysis module uses the ICEEMDAN modal decomposition algorithm, a consistent random sampling strategy and a sample entropy fast evaluation method, and combines variational modal decomposition and the Autoformer model to predict mine microseisms and microseisms in multiple steps. The regional difference analysis module uses the short-time Fourier transform, adaptive wavelet analysis technology and high-order difference hybrid numerical methods to simulate and analyze the regional characteristic differences of mine seismic wave propagation. The intelligent data processing module integrates multi-source geophysical data, real-time dynamic imaging data, microseismic prediction data and seismic wave propagation spatial difference data, and uses a deep learning intelligent risk assessment model, a dynamic weight ensemble learning method and an attention mechanism to conduct real-time dynamic assessment and automatic early warning of concealed structures and mine water disaster risks; the input-output relationships between the modules are specifically as follows:

[0006] The input of the multi-source data comprehensive detection module is GPS positioning data, geophysical exploration data and electromagnetic interference data, and the output is a high-precision three-dimensional geological structure model;

[0007] The input of the real-time imaging module is the three-dimensional geological structure model and resistivity monitoring data, and the output is a real-time underground resistivity imaging image, which finely displays the underground concealed structure area;

[0008] The input of the microseismic multi-scale intelligent analysis module is the original mine microseismic monitoring data, and the output is multi-scale prediction data of mine microseismic events, including the energy, occurrence frequency and spatial position of microseismic events;

[0009] The input of the seismic wave signal recognition module is the mine seismic wave propagation signal collected in real time, and the output is the recognized seismic wave characteristic signal, including the earthquake source position, magnitude and propagation path characteristics;

[0010] The input of the regional difference analysis module is the seismic wave characteristic signal output by the seismic wave signal recognition module and the geological parameters of each region, and the output is the spatial difference characteristic data of the seismic wave propagation path;

[0011] The input of the intelligent data processing module is the three-dimensional geological structure model, the real-time imaging image, the microseismic prediction data and the seismic wave propagation spatial difference data, and the output is the real-time dynamic assessment result of the concealed structure and mine water disaster risk and the automatic early warning information.

[0012] As a further solution of the present invention, the multi-source data comprehensive detection module obtains GPS positioning data, geophysical exploration data, and electromagnetic interference data in real time. Based on the adaptive spatial interpolation technology, by automatically detecting the local errors of the three-dimensional geological model underground in the mine in real time, the resolution of the interpolation grid is dynamically updated, and the missing data area in the three-dimensional geological model is repaired quickly in real time.

[0013] As a further solution of the present invention, in the multi-source data comprehensive detection module, the local errors of the three-dimensional geological model include but are not limited to the first local error caused by the missing data acquisition due to sensor failure, and the second local error caused by the local geological disturbance caused by encountering complex geological structures and mining activities, resulting in data loss. For the local error caused by the missing data acquisition due to sensor failure, the first error is repaired by real-time monitoring of the sensor status and automatically enabling the redundant interpolation method of fusing the data of adjacent sensors within a set range. The second error is repaired by real-time detecting the geological structure disturbance area and establishing a targeted interpolation model using the disturbance characteristics.

[0014] As a further solution of the present invention, in the multi-source data comprehensive detection module, during the repair process of the first error, taking the faulty sensor as the center, the data of the four to six nearest sensors are selected for weighted fusion interpolation; during the repair process of the second error, a fault fracture feature matching interpolation model, a goaf stress release dynamic interpolation model, and a groundwater seepage feature interpolation model are respectively established according to the spatial distribution characteristics of faults, fracture zones, goafs, and underground water bodies.

[0015] As a further solution of the present invention, during the process of the multi-source data comprehensive detection module repairing the second error:

[0016] The fault fracture feature matching interpolation model realizes the refined repair of the spatial trend of the data missing area by identifying the spatial distribution characteristics, fracture density, and extension direction of the fault fracture zone, and using the anisotropic interpolation algorithm. Its repair formula is:

[0017]

[0018] In the formula: x and y are respectively the abscissa and ordinate of the position to be interpolated, and the reference coordinate system takes the set reference point in the detection area as the origin. The positive direction of the x-axis points to the extension direction of the main roadway in the mine, and the positive direction of the y-axis is perpendicular to the main roadway and extends to the right. Z(x, y) is the value at the position to be inserted in the fault fracture feature matching interpolation model, i is the index of the known data point, n is the number of known data points, and λ i (x, y) is the anisotropic weight coefficient, which is obtained by calculating the cosine similarity between the fracture density and the fracture direction vector. θ iis the angle between the interpolation point to be interpolated and the fracture direction of the i-th known data point, D i is the fracture density of the i-th known data point, k is the index variable in the summation process, θ k is the angle between the k-th interpolation point to be interpolated and the fracture direction of the known data point, D k is the fracture density of the k-th known data point, Z(x i ,y i ) is the observed value at the i-th known data point;

[0019] The goaf stress release dynamic interpolation model is based on the stress change of the surrounding rock strata of the goaf monitored in real time, establishes a dynamic interpolation function, reflects the change of the geological disturbance area caused by stress release in real time, and realizes accurate data supplementation. Its repair formula is:

[0020] Q(x, y, t) = Q0(x, y) + β·ΔT(x, y, t)

[0021] Where: t is the current time, Q(x, y, t) is the interpolation at the current position (x, y) and the current time t in the goaf stress release dynamic interpolation model, Q0(x, y) is the initial state data, β is the stress sensitivity coefficient obtained from the properties of the rock strata material, and ΔT(x, y, t) is the real-time monitored rock strata stress change value at the current time t;

[0022] The interpolation model of the underground water body seepage characteristics analyzes the dynamic monitoring data of the underground water body seepage path, flow velocity and underground water level, adopts the flow field-oriented interpolation technology, and supplements the missing data in the groundwater influence area in real time. Its repair formula is:

[0023]

[0024] In the formula: W(x, y, t) is the underground water level to be interpolated at time t, x0, y0, t0 are the abscissa, ordinate and initial time of the initial water level monitoring point, W0(x0, y0, t0) is the initial water level, γ is the underground water flow field orientation factor determined by the real-time flow field monitoring data, τ is the integration variable, and U(x, y, τ) is the underground water flow velocity, is the underground water level gradient.

[0025] As a further solution of the present invention, the real-time imaging module performs differential processing on the resistivity data sequence collected at consecutive moments based on the dynamic adaptive grid refinement technology, calculates the resistivity change gradient at adjacent moments, identifies the resistivity anomaly trend area, and determines the significance of the resistivity anomaly in real time by setting a gradient threshold. The adaptive differential gradient method is used to accurately identify the boundary position of the resistivity anomaly area, and based on the gradient value magnitude and spatial distribution range of the anomaly area, the resolution of the grid is dynamically adjusted. The grid size is automatically refined in the area where the gradient value is within the set first-level threshold range, and the grid size is automatically enlarged in the area where the gradient value is within the set second-level threshold range. The grid resolution and refinement area are dynamically adjusted and optimized in real time to finely capture and identify hidden structures in real time.

[0026] As a further solution of the present invention, the seismic wave signal recognition module acquires the real-time collected mine seismic wave propagation signal, identifies the source location, magnitude, and propagation path characteristics of the seismic wave, and transmits the identified characteristics to the regional difference analysis module.

[0027] As a further solution of the present invention, the microseismic multi-scale intelligent analysis module dynamically determines the frequency and amplitude of threshold update by calculating the sample entropy change trend of the data sequence within the sliding window in real time, and uses the real-time sample entropy threshold to guide the consistent random sampling strategy. When the sample entropy exceeds the current dynamic threshold, the distance between the sampling bands is shortened to increase the local sampling density, and the number of sampling points in this area is increased by 20% - 30% additionally. When the sample entropy is lower than the current dynamic threshold, the distance between the sampling points is extended, and the number of sampling points is reduced to 60% - 80% of the original design value, adaptively and dynamically adjusting the spatial distribution and density of the sampling points.

[0028] As a further solution of the present invention, the regional difference analysis module acquires the rock elastic modulus, Poisson's ratio, density, and medium damping parameters of different areas of the mine and the real-time seismic wave propagation data in real time, establishes the geological characteristic model of each area, uses the real-time seismic wave propagation data to solve the spatial partial derivative of the seismic wave data collected at consecutive moments, obtains the wave field spatial gradient distribution, analyzes the local wave field error in real time based on the wave field spatial gradient, sets the dynamic adjustment threshold, and evaluates the local Courant stability condition in real time by analyzing the local wave velocity difference and the seismic waveform frequency change trend in real time. The upper limit of the time step is automatically calculated from the real-time seismic wave velocity data, and the optimal time step that meets the stability requirements is dynamically selected to accurately simulate the spatial differentiation characteristics of the seismic wave propagation path.

[0029] As a further solution of the present invention, the intelligent data processing module monitors the change rate of the input data features in real time by using the sliding window method, calculates the abnormality degree of the features by using the exponentially weighted moving average method with real-time update, adjusts the update frequency of the spatial weight matrix and the distribution density of the weights in real time based on the current abnormality degree, realizes the hierarchical interaction between local features and global features through the hierarchical attention mechanism, the local attention layer quickly responds to the position where the abnormality appears, the global attention layer dynamically coordinates the attention intensity of each local feature, and triggers the adaptive attention adjustment algorithm to quickly focus on the current most critical risk area through the data of the high-risk area of the hidden structure fed back in real time.

[0030] Technical effects of a hidden structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to the present invention:

[0031] Through multi-scale data fusion and real-time dynamic analysis technology, the present invention comprehensively integrates differential GPS positioning, Kalman filtering, adaptive noise reduction, Kriging interpolation, joint inversion, adaptive non-linear regularization inversion, GPU parallel computing, ICEEMDAN modal decomposition, consistent random sampling strategy, variational modal decomposition and Autoformer deep learning model, as well as short-time Fourier transform, adaptive wavelet analysis, high-order difference hybrid numerical method and dynamic weight integrated learning and attention mechanism of intelligent data processing, realizes accurate detection of hidden structures and real-time dynamic prediction and early warning of mine water disasters, significantly reduces casualties and economic losses, and ensures the safe production of mines and the construction of smart mines. Description of the drawings

[0032] Figure 1 It is the system block diagram of the present invention;

[0033] Figure 2 It is the flow chart of modal decomposition of the present invention;

[0034] Figure 3 It is the process of constructing the input-output data set of multiple features of the present invention;

[0035] Figure 4 It is the architecture diagram of the Autoformer model of the present invention. Detailed implementation manners

[0036] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described content is only a part of the present invention, rather than all of it. All other technical solutions obtained by those of ordinary skill in the art based on the content of the present invention without making creative efforts belong to the scope of protection of the present invention.

[0037] Such as Figure 1As shown in the figure, a concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters proposed by the present invention includes a multi-source data comprehensive detection module, a real-time imaging module, a microseismic multi-scale intelligent analysis module, a seismic wave signal recognition module, a regional difference analysis module and an intelligent data processing module. The multi-source data comprehensive detection module uses differential GPS positioning technology, integrates Kalman filtering and adaptive noise reduction algorithms, and constructs a three-dimensional geological structure model by using Kriging interpolation and joint inversion methods. The real-time imaging module constructs a real-time underground structure imaging through an adaptive non-linear regularization inversion method and accelerates the imaging process by using GPU parallel computing. The microseismic multi-scale intelligent analysis module uses the ICEEMDAN modal decomposition algorithm, a consistent random sampling strategy and a sample entropy fast evaluation method, combines variational mode decomposition and the Autoformer model to predict mine microseisms at multiple time steps. The regional difference analysis module uses short-time Fourier transform, adaptive wavelet analysis technology and a high-order difference hybrid numerical method to simulate and analyze the regional characteristic differences of mine seismic wave propagation. The intelligent data processing module integrates multi-source geophysical data, real-time dynamic imaging data, microseismic prediction data and seismic wave propagation spatial difference data, and uses a deep learning intelligent risk assessment model, a dynamic weight ensemble learning method and an attention mechanism to perform real-time dynamic assessment and automatic early warning on the concealed structure and mine water disaster risks;

[0038] As Figure 2 Shown is the flow chart of modal decomposition in the microseismic multi-scale intelligent analysis module. Before constructing the data set, the original data is first preprocessed. By performing secondary modal decomposition on the preprocessed data set, the signal can be analyzed at multiple scales and detailed information in different frequency bands can be extracted, thereby obtaining several characteristic modal functions. Subsequently, these characteristic modal functions are combined with the time series data of the original data set to finally obtain a multi-source characteristic time series sample set. Then as Figure 3 Shown, a sliding window method is used to construct the input-output data set of multi-source characteristics. The sliding window method divides the data into multiple subsequences by sliding a window of fixed length on the time series to form sample pairs. These subsequences serve as the input of the model and, combined with the corresponding output labels, constitute the final data set. A multi-source characteristic input-single output sample data set is constructed using the multi-source characteristic time series sample set.

[0039] First, perform ICEEMDAN decomposition on the original time series data. Through multiple noise addition and noise adjustment, ICEEMDAN makes the decomposition result more robust to noise, and adaptively adjusts the noise and mode extraction process, thus reducing the mode mixing phenomenon and making each mode have a more distinct frequency distribution. After ICEEMDAN decomposition, several C-IMF sequences are obtained. Then, using the method of data selection sampling, uniformly random sampling is performed on each C-IMF sequence, and the sample entropy of the sampled sequence is calculated. Generally, the larger the sample entropy, the more complex, disordered, and higher the frequency components of the signal; on the contrary, a sequence with a smaller sample entropy represents a relatively stable signal, lower frequency components, or a stronger trend. According to the size of the sample entropy, the C-IMF sequences are reconstructed into three sequences: high-frequency, medium-frequency, and low-frequency. Subsequently, VMD decomposition is performed on the reconstructed high-frequency, medium-frequency, and low-frequency sequences. The decomposition effect of VMD highly depends on the selection of the number of modes. When the number of modes is too small, the VMD algorithm may act as an adaptive filter bank, resulting in the filtering out of key information in the original signal and thus reducing the prediction accuracy; while when the number of modes is too large, the central frequencies of adjacent mode components are close, which may lead to the generation of redundancy or additional noise. In this paper, the decomposition number K is initially estimated based on the sample entropy of each sequence, and the appropriate decomposition number is selected according to the central frequency distribution under different decomposition numbers. Finally, n V-IMF sequences are obtained by decomposition. Combining these V-IMF sequences with the original data features, n + 1 multivariate feature time series sample sets are obtained. Then, the sliding window method is used to construct the dataset. The specific method is as follows: First, set the time window Δt (i.e., the time step) and define the sliding step S. Within each sliding window, the first Δt time series values together with the corresponding V-IMF component data are used as the input features of the model; subsequently, at the end of each time window, the data values at the next moment or multiple subsequent moments are used as the output. Through this process, the model will learn how to predict one or more future time series values based on the features of the first Δt time steps.

[0040] Such as Figure 4The figure shows the architecture diagram of the Autoformer model used in the present invention. Specifically, the model decomposes the input time series into a trend component and a seasonal component. This decomposition process uses a sliding window for smoothing to reduce the interference of noise on the model. The input features of the encoder and decoder are embedded into a high-dimensional space. Each encoder module consists of multiple encoding layers and is responsible for extracting time series features. Each layer of the encoder includes an autocorrelation layer, a feed-forward neural network, a LayerNormalization layer, and a Dropout layer. The autocorrelation layer is used to capture the long-term dependencies in the time series. The feed-forward neural network consists of two fully connected networks and incorporates a non-linear activation function to enhance the feature representation ability. The LayerNormalization layer and the Dropout layer help to stabilize the training process and prevent overfitting. Each decoder module also consists of multiple decoding layers, where each layer includes the interaction between the autocorrelation layer and the cross-autocorrelation layer, the interaction between the decoder layer and the encoder layer, and the reconstruction of the trend and seasonal components. In the long time series prediction task, the decoder output is combined with the encoder features to reconstruct the trend component and the seasonal component, and finally generate the prediction result.

[0041] The specific input-output relationships between the modules are as follows:

[0042] The input of the multi-source data integrated detection module is GPS positioning data, geophysical exploration data, and electromagnetic interference data, and the output is a high-precision three-dimensional geological structure model;

[0043] The input of the real-time imaging module is the three-dimensional geological structure model and resistivity monitoring data, and the output is a real-time underground resistivity imaging image, which finely displays the underground buried structure area;

[0044] The input of the microseismic multi-scale intelligent analysis module is the original data of mine microseismic monitoring, and the output is the multi-scale prediction data of mine microseismic events, including the energy, occurrence frequency, and spatial location of microseismic events;

[0045] The input of the seismic wave signal recognition module is the seismic wave propagation signal collected in real time in the mine, and the output is the recognized seismic wave characteristic signal, including the earthquake source location, magnitude, and propagation path characteristics;

[0046] The input of the regional difference analysis module is the seismic wave characteristic signal output by the seismic wave signal recognition module and the geological parameters of each region, and the output is the spatial difference characteristic data of the seismic wave propagation path;

[0047] The input of the intelligent data processing module is the three-dimensional geological structure model, the real-time imaging image, the microseismic prediction data, and the spatial difference data of seismic wave propagation, and the output is the real-time dynamic assessment result of the buried structure and the risk of mine water disaster and the automatic warning information.

[0048] To solve the technical problems raised in the background art, the working process of a concealed structure detection system for intelligent monitoring, early warning and prevention of mine water hazards proposed by the present invention is as follows:

[0049] (1) Multi-source data comprehensive detection module: Using differential GPS positioning technology, integrating Kalman filtering and adaptive noise reduction algorithms, processing the input GPS positioning data, geophysical exploration data, and electromagnetic interference data, and constructing a high-precision three-dimensional geological structure model by using Kriging interpolation and joint inversion methods;

[0050] (2) Real-time imaging module: Taking the three-dimensional geological structure model and real-time monitored resistivity data as input, constructing a real-time underground resistivity imaging image through an adaptive non-linear regularization inversion method, and accelerating the imaging process by using GPU parallel computing to finely identify and display the underground concealed structure area;

[0051] (3) Microseismic multi-scale intelligent analysis module: Taking the original mine microseismic monitoring data as input, using the ICEEMDAN modal decomposition algorithm, consistent random sampling strategy, and sample entropy rapid evaluation method, combining variational mode decomposition and Autoformer deep learning model, realizing multi-scale prediction of mine seismic and microseismic events, and outputting microseismic event energy, occurrence frequency, and spatial position information;

[0052] (4) Seismic wave signal recognition module: Inputting the real-time collected mine seismic wave propagation signals, outputting the recognized seismic wave characteristic signals through signal processing and feature extraction, including the source location, magnitude, and propagation path characteristics;

[0053] (5) Regional difference analysis module: Taking the seismic wave characteristic signals output by the seismic wave signal recognition module and the geological parameters of each region as input, using the short-time Fourier transform, adaptive wavelet analysis, and high-order difference hybrid numerical methods to simulate and analyze the regional characteristic differences of seismic wave propagation in the mine, and outputting the spatial difference characteristic data of the seismic wave propagation path;

[0054] (6) Intelligent data processing module: Integrating multi-source geophysical data, real-time dynamic imaging data, microseismic prediction data, and seismic wave propagation spatial difference data as input, using a deep learning intelligent risk assessment model, dynamic weight ensemble learning method, and attention mechanism to realize real-time dynamic assessment and automatic early warning of concealed structures and mine water hazard risks, and outputting risk assessment results and early warning information.

[0055] It should be noted that the multi-source data comprehensive detection module obtains GPS positioning data, geophysical exploration data, and electromagnetic interference data in real time. Based on the adaptive space interpolation technology, by automatically detecting the local errors of the three-dimensional geological model of the mine underground in real time, dynamically updating the interpolation grid resolution, and quickly repairing the missing data area in the three-dimensional geological model in real time.

[0056] It should be further noted that in the multi-source data comprehensive detection module, the local errors of the three-dimensional geological model include but are not limited to the first local error caused by the missing data acquisition due to sensor failures, and the second local error caused by the missing data due to local geological disturbances caused by complex geological structures and mining activities. For the local error caused by the missing data acquisition due to sensor failures, the first error is repaired by real-time monitoring of the sensor status and automatically enabling the redundant interpolation method of fusing the data of adjacent sensors within a set range. The second error is repaired by real-time detecting the disturbed area of the geological structure and establishing a targeted interpolation model using the disturbance characteristics.

[0057] It should be further noted that in the multi-source data comprehensive detection module, during the repair process of the first error, the data of the four to six closest sensors are selected for weighted fusion interpolation with the faulty sensor as the center. During the repair process of the second error, a fault fracture feature matching interpolation model, a goaf stress release dynamic interpolation model, and a groundwater seepage feature interpolation model are respectively established according to the spatial distribution characteristics of faults, fracture zones, goafs, and underground water bodies.

[0058] It should be further noted in detail that during the process of the multi-source data comprehensive detection module repairing the second error:

[0059] The fault fracture feature matching interpolation model realizes the refined repair of the spatial trend of the data missing area by identifying the spatial distribution characteristics, fracture density, and extension direction of the fault fracture zone, and uses the anisotropic interpolation algorithm. Its repair formula is:

[0060]

[0061] In the formula: x and y are respectively the abscissa and ordinate of the position to be interpolated. The reference coordinate system takes the set reference point in the detection area as the origin. The positive direction of the x-axis points to the extension direction of the main roadway in the mine, and the positive direction of the y-axis is perpendicular to the main roadway and extends to the right. Z(x, y) is the value at the position to be inserted in the fault fracture feature matching interpolation model. i is the index of the known data point, n is the number of known data points, λ i (x, y) is the anisotropic weight coefficient, which is obtained by calculating the cosine similarity between the fracture density and the fracture direction vector. θ i is the angle between the direction of the fracture at the point to be interpolated and the direction of the fracture of the i-th known data point. D i is the fracture density of the i-th known data point. k is the index variable in the summation process. θ k is the angle between the direction of the fracture at the k-th point to be interpolated and the direction of the fracture of the known data point. D k is the fracture density of the k-th known data point. Z(x i ,y i) is the observed value at the i-th known data point;

[0062] The goaf stress release dynamic interpolation model is based on the stress change of the surrounding rock strata of the goaf monitored in real time, establishes a dynamic interpolation function, reflects the change of the geological disturbance area caused by stress release in real time, and realizes accurate data supplementation. Its repair formula is:

[0063] Q(x, y, t) = Q0(x, y) + β·ΔT(x, y, t)

[0064] Where: t is the current time, Q(x, y, t) is the interpolation at the current position (x, y) at the current time t in the goaf stress release dynamic interpolation model, Q0(x, y) is the initial state data, β is the stress sensitivity coefficient obtained from the properties of the rock strata material, and ΔT(x, y, t) is the real-time monitored stress change value of the rock strata at the current time t;

[0065] The interpolation model of the seepage characteristics of underground water bodies analyzes the dynamic monitoring data of the seepage path, flow velocity and groundwater level of underground water bodies, adopts the flow field-oriented interpolation technology, and supplements the missing data in the groundwater influence area in real time. Its repair formula is:

[0066]

[0067] In the formula: W(x, y, t) is the groundwater level to be interpolated at time t, x0, y0, and t0 are the abscissa, ordinate and initial time of the initial water level monitoring point, W0(x0, y0, t0) is the initial water level, γ is the groundwater flow field orientation factor determined by the real-time flow field monitoring data, τ is the integration variable, U(x, y, τ) is the groundwater flow velocity, is the groundwater level gradient.

[0068] The multi-source data comprehensive detection module obtains GPS positioning data, geophysical exploration data, and electromagnetic interference data in real time. Based on the adaptive spatial interpolation technology, it dynamically detects and repairs the local errors of the three-dimensional underground geological model of the mine. During the specific implementation process, for example, when conducting detection in a certain mine area, the system monitors in real time that the sensor numbered S10 fails, resulting in data loss. The system automatically selects the data of six sensors, namely S7, S8, S9, S11, S12, and S13, which are the closest to it, and uses the weighted fusion interpolation method to repair the missing area, so that the integrity of the model data quickly recovers from 85% to 100%. At the same time, when the detection area encounters a complex fault fracture zone, the spatial position and density change of the fracture zone are determined in real time through geophysical exploration, and the anisotropic interpolation method is used to repair the data missing area. In the example, the initial fracture density at the position (50, 100) of the fracture zone is 0.8, and the direction angle is 30°. Through model calculation, the accuracy of real-time interpolation repair reaches more than 0.95. In the goaf area, taking a goaf at the position (120, 150) as an example, the initial rock stratum stress data is 5 MPa. By real-time monitoring, the rock stratum stress changes to 0.8 MPa, and after calculation using the dynamic interpolation model, the interpolation data is accurately updated to 5.64 MPa. For the underground water seepage area, the initial water level at the initial monitoring position (200, 250) is 10 m. By real-time monitoring, the flow velocity is 0.05 m / s, the water level gradient is 0.002 m / m, and the flow field guiding factor is determined to be 0.95. After real-time calculation and interpolation repair, the accuracy of the water level data is increased to more than 99%.

[0069] The present invention obtains GPS positioning data, geophysical exploration data, and electromagnetic interference data in real time through the multi-source data comprehensive detection module. Based on the adaptive spatial interpolation technology, it dynamically detects and repairs the local errors of the three-dimensional underground geological model of the mine. Among them, redundant data interpolation and feature matching interpolation methods are respectively adopted for sensor failures and complex geological structure disturbances, including fault fracture feature matching, goaf stress release dynamic interpolation, and underground water seepage feature interpolation models, realizing the real-time refined supplement of data missing areas, effectively improving the detection accuracy of hidden structures, significantly enhancing the real-time prediction and early warning ability of mine water disasters, greatly reducing the mine safety risk, and ensuring the safe production of the mine.

[0070] It should be noted that the real-time imaging module is based on the dynamic adaptive grid refinement technology. It performs differential processing on the resistivity data sequences collected at consecutive moments, calculates the resistivity change gradient at adjacent moments, identifies the resistivity anomaly trend regions, and determines the significance of resistivity anomalies in real time by setting a gradient threshold. It uses the adaptive differential gradient method to accurately identify the boundary positions of the resistivity anomaly regions, and based on the magnitude of the gradient values and the spatial distribution range of the anomaly regions, dynamically adjusts the resolution of the grid. The grid size is automatically refined in the regions where the gradient values are within the set first-level threshold range, and the grid size is automatically enlarged in the regions where the gradient values are within the set second-level threshold range, dynamically adjusting and optimizing the grid resolution and refinement regions in real time to finely capture and identify hidden structures in real time.

[0071] In a specific implementation example, real-time monitoring is carried out in the main roadway area of the mine, and resistivity data at two consecutive moments are collected. Among them, the resistivity at the regional position (80, 120) drops from the initial 20 Ω·m to 15 Ω·m, and the resistivity gradient change reaches 0.5 Ω·m / m, exceeding the set first-level threshold of 0.4 Ω·m / m. The system automatically refines the grid size in this area to 50% of the original size, increasing the grid accuracy from the original 1 m to 0.5 m, effectively improving the identification accuracy of the anomaly area. At the same time, the resistivity gradient change at the regional position (180, 200) is only 0.2 Ω·m / m, lower than the second-level threshold of 0.3 Ω·m / m. The system automatically enlarges the grid size to 150% of the original size, that is, 1.5 m, reducing the waste of unnecessary computing resources. Through the above dynamic adjustment, hidden structure areas in the mine are finely captured and identified in real time, the detection accuracy is increased to over 98%, the mine safety risk prediction and early warning ability is significantly enhanced, and the probability of on-site safety accidents is reduced by more than 30%.

[0072] The real-time imaging module adopts the dynamic adaptive grid refinement technology. By performing real-time differential processing on the resistivity data sequences and calculating the change gradient, it accurately identifies the resistivity anomaly regions and their boundary positions, and dynamically adjusts the grid resolution in real time according to the gradient values and spatial distribution, realizing efficient, accurate capture and real-time fine imaging of hidden structures, effectively improving the detection accuracy of hidden structures, enhancing the mine water hazard prediction and early warning ability, and significantly reducing the mine safety risks and the probability of accidents.

[0073] It should be noted that the seismic wave signal identification module acquires the real-time collected mine seismic wave propagation signals, identifies the source position, magnitude, and propagation path characteristics of the seismic waves, and transmits the identified characteristics to the regional difference analysis module.

[0074] It should be noted that the microseismic multi-scale intelligent analysis module dynamically determines the frequency and amplitude of threshold update by calculating the change trend of sample entropy of the data sequence within the sliding window in real time, and uses the real-time sample entropy threshold to guide the consistent random sampling strategy. When the sample entropy exceeds the current dynamic threshold, the distance between the sampling bands is shortened to increase the local sampling density, and the number of sampling points in this area is increased by 20% - 30% additionally. When the sample entropy is lower than the current dynamic threshold, the distance between the sampling points is extended, and the number of sampling points is reduced to 60% - 80% of the original design value, adaptively and dynamically adjusting the spatial distribution and density of the sampling points.

[0075] The microseismic multi-scale intelligent analysis module dynamically determines the frequency and amplitude of threshold update by calculating the change trend of sample entropy of the data sequence within the sliding window in real time, and uses the real-time sample entropy threshold to guide the consistent random sampling strategy. The specific implementation process is as follows: For example, the length of the continuously monitored microseismic data sequence in a certain mine area is 1000 data points, and the initial designed sampling point density is to sample once every 10 data points. Through real-time sample entropy calculation, it is found that when the sequence position is in the section from 200 to 300, the sample entropy rapidly increases from the initial threshold of 0.5 to 0.75, exceeding the dynamically updated threshold of 0.7. At this time, the sampling strategy is immediately adjusted, and the sampling density, which was originally once every 10 data points, is increased to once every 7 data points, which is equivalent to an additional increase of about 30% of the sampling points in this section to more precisely capture the change trend of microseismic events. When the sequence position is in the section from 700 to 800, the sample entropy decreases from 0.5 to 0.3, lower than the dynamically updated threshold of 0.4. The system automatically reduces the sampling point density to 70% of the original design value, that is, to sample once every about 14 data points, thus saving computing resources and maintaining a high analysis accuracy. This method of adaptively and dynamically adjusting the sampling point density realizes the fine dynamic monitoring of microseismic data and efficient data processing.

[0076] It should be noted that the regional difference analysis module obtains the elastic modulus, Poisson's ratio, density and medium damping parameters of rocks in different areas of the mine and the real-time propagation data of seismic waves in real time, establishes the geological feature models of each area, uses the real-time propagation data of seismic waves to solve the spatial partial derivatives of the seismic wave data collected at consecutive moments, obtains the wave field spatial gradient distribution, analyzes the local wave field error in real time according to the wave field spatial gradient, sets the dynamic adjustment threshold, evaluates the local Courant stability condition in real time by analyzing the change trend of local wave velocity difference and seismic waveform frequency, automatically calculates the upper limit of the time step according to the real-time seismic wave velocity data, and dynamically selects the optimal time step that meets the stability requirements to accurately simulate the spatial differentiation characteristics of the seismic wave propagation path.

[0077] In a specific implementation example, during the detection process of a certain mine, the elastic modulus of rocks measured in real time in areas A, B, and C are 15 GPa, 18 GPa, and 12 GPa respectively, the Poisson's ratios are 0.25, 0.28, and 0.22 respectively, and the densities are 2600 kg / m 3 , 2700 kg / m 3 , and 2500 kg / m 3 respectively. The medium damping parameters are 0.02, 0.015, and 0.025 respectively. The system continuously collects seismic wave data propagating in different areas, solves the spatial partial derivatives in real time, and calculates that the average value of the seismic wave spatial gradient in area A is 0.35, in area B is 0.45, and in area C is 0.25. Based on this, the local wave field error of each area is evaluated in real time and the dynamic adjustment threshold is set. For example, when the wave field gradient in area B exceeds the set threshold of 0.4, the system automatically increases the spatial difference order to improve the calculation accuracy. At the same time, according to the real-time seismic wave velocity differences in different areas, such as the wave velocity in area A is 3200 m / s, in area B is 3500 m / s, and in area C is 3000 m / s, the change trend of the seismic waveform frequency is analyzed in real time and the Courant stability condition is automatically calculated, and the optimal time step that meets the stability requirements is dynamically selected. The time step in area A is determined to be 0.0002 s, in area B is 0.00018 s, and in area C is 0.00022 s, so as to achieve high-precision real-time simulation of the spatial differentiation characteristics of the seismic wave propagation path, reduce the overall wave field error to within 3%, and significantly improve the ability to identify hidden structures and predict mine safety risks.

[0078] It should be noted that the intelligent data processing module uses the sliding window method to monitor the change rate of the input data characteristics in real time, calculates the abnormal degree of the characteristics by using the exponentially weighted moving average method with real-time update, adjusts the update frequency of the spatial weight matrix and the distribution density of the weights in real time based on the current abnormal degree, and realizes the hierarchical interaction between local features and global features through the hierarchical attention mechanism. The local attention layer quickly responds to the position where the anomaly appears, and the global attention layer dynamically coordinates the attention intensity of each local feature. Through the real-time feedback data of the high-risk areas of hidden structures, the adaptive attention adjustment algorithm is triggered to quickly focus on the current most critical risk area.

[0079] The intelligent data processing module monitors the change rate of input data features in real time by using the sliding window method, calculates the degree of abnormality of the features by using the exponentially weighted moving average method with real-time update, and adjusts the update frequency of the spatial weight matrix and the distribution density of the weights in real time based on the current degree of abnormality. During the specific execution process, for example, in a hidden structure detection task in a certain mine, the system detects that the exponentially weighted moving average (EWMA) of the degree of abnormality of the mine data rises from 0.2 to 0.6 within the time window T1 (0 - 10 minutes), indicating that the data features have changed drastically. At this time, the system automatically increases the update frequency of the spatial weight matrix to update once every 5 seconds and increases the weight distribution density of the abnormal area by 50% to make it receive higher attention. Within the time window T2 (10 - 20 minutes), when the EWMA drops to 0.3, the system reduces the update frequency to once every 10 seconds and moderately increases the weight distribution density of the normal area by 15% to ensure the overall monitoring coverage. The system also adopts a hierarchical attention mechanism for local and global feature interaction. The response time of the local attention layer to abnormal data is shortened to 1.5 seconds to ensure that high-risk areas are given computing resources first; the global attention layer dynamically coordinates the features of multiple areas of the mine and comprehensively analyzes the evolution trend of the hidden structure risk. Finally, at the time window T3 (20 - 30 minutes), the system locks the current most critical hidden structure high-risk area through the adaptive attention adjustment algorithm and triggers the warning mechanism, predicting the specific area where a mine tremor may occur 20 minutes in advance, successfully avoiding a potential mine safety accident and improving the accuracy of hidden structure detection and the response speed of mine water disaster warning.

[0080] In summary, the present invention realizes the precise detection of hidden structures and the real-time dynamic prediction and warning of mine water disasters through multi-scale data fusion and real-time dynamic analysis technologies, integrating differential GPS positioning, Kalman filtering, adaptive noise reduction, Kriging interpolation, joint inversion, adaptive non-linear regularization inversion, GPU parallel computing, ICEEMDAN modal decomposition, consistent random sampling strategy, variational modal decomposition, and the Autoformer deep learning model, as well as short-time Fourier transform, adaptive wavelet analysis, high-order difference hybrid numerical methods, and the dynamic weight integrated learning and attention mechanism of intelligent data processing, significantly reducing casualties and economic losses and ensuring the safe production of mines and the construction of smart mines.

[0081] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claimed rights.

[0082] Finally: The above are only embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An undetected structure detection system for intelligent monitoring, early warning and prevention of mine water disasters, characterized in that, It includes a multi-source data comprehensive detection module, a real-time imaging module, a microseismic multi-scale intelligent analysis module, a seismic wave signal recognition module, a regional difference analysis module, and an intelligent data processing module. The multi-source data comprehensive detection module uses differential GPS positioning technology, integrates Kalman filtering and adaptive noise reduction algorithms, and constructs a three-dimensional geological structure model using Kriging interpolation and joint inversion methods. The real-time imaging module constructs a real-time underground structure imaging through an adaptive non-linear regularization inversion method and accelerates the imaging process using GPU parallel computing. The microseismic multi-scale intelligent analysis module uses the ICEEMDAN modal decomposition algorithm, a consistent random sampling strategy, and a sample entropy fast evaluation method, and combines variational mode decomposition and the Autoformer model for multi-step prediction of mine tremors and microseisms. The regional difference analysis module uses short-time Fourier transform, adaptive wavelet analysis technology, and a high-order difference hybrid numerical method to simulate and analyze the regional characteristic differences of mine seismic wave propagation. The intelligent data processing module integrates multi-source geophysical data, real-time dynamic imaging data, microseismic prediction data, and seismic wave propagation spatial difference data, and uses a deep learning intelligent risk assessment model, a dynamic weight ensemble learning method, and an attention mechanism to perform real-time dynamic assessment and automatic early warning of hidden structures and mine water disaster risks.

2. The concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 1, wherein The multi-source data comprehensive detection module obtains GPS positioning data, geophysical exploration data, and electromagnetic interference data in real time. Based on adaptive spatial interpolation technology, it dynamically updates the interpolation grid resolution by automatically detecting local errors in the three-dimensional geological model of the mine underground in real time, and quickly repairs the missing data area in the three-dimensional geological model in real time.

3. The concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 2, wherein, In the multi-source data comprehensive detection module, the local errors of the three-dimensional geological model include, but are not limited to, the first local error caused by missing data acquisition due to sensor failures, and the second local error caused by missing data due to local geological disturbances caused by complex geological structures and mining activities. For the local error caused by missing data acquisition due to sensor failures, the first error is repaired by real-time monitoring of the sensor status and automatically enabling the redundant interpolation method of fusing data from adjacent sensors within a set range. The second error is repaired by real-time detecting the disturbed area of the geological structure and establishing a targeted interpolation model using the disturbance characteristics.

4. The concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 3, characterized in that, In the multi-source data comprehensive detection module, during the repair process of the first error, with the faulty sensor as the center, the data of four to six sensors closest in distance are selected for weighted fusion interpolation; during the repair process of the second error, a fault fracture feature matching interpolation model, a goaf stress release dynamic interpolation model, and a groundwater seepage feature interpolation model are established respectively according to the spatial distribution characteristics of faults, fracture zones, goafs, and underground water bodies.

5. The concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 4, characterized in that, During the process of the multi-source data comprehensive detection module repairing the second error: The fault fracture feature matching interpolation model realizes the refined repair of the spatial trend of the data missing area by identifying the spatial distribution characteristics, fracture density, and extension direction of the fault fracture zone and using the anisotropic interpolation algorithm; The goaf stress release dynamic interpolation model is based on the stress changes of the surrounding rock strata of the goaf monitored in real time, establishes a dynamic interpolation function, reflects the changes in the geological disturbance area caused by stress release in real time, and realizes accurate data supplementation; The underground water seepage characteristics interpolation model analyzes the dynamic monitoring data of the underground water seepage path, flow velocity and underground water level, and uses the flow field-oriented interpolation technology to supplement the data missing in the groundwater influence area in real time.

6. The concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 1, characterized in that, The real-time imaging module is based on the dynamic adaptive grid refinement technology, performs differential processing on the resistivity data sequence collected at consecutive moments, calculates the resistivity change gradient at adjacent moments, identifies the resistivity anomaly trend area, and judges the significance of the resistivity anomaly in real time by setting the gradient threshold. The adaptive differential gradient method is used to accurately identify the boundary position of the resistivity anomaly area. Based on the magnitude and spatial distribution range of the gradient value in the anomaly area, the resolution of the grid is dynamically adjusted. The grid size in the area where the gradient value is within the set first-level threshold range is automatically refined, and the grid size in the area where the gradient value is within the set second-level threshold range is automatically enlarged. The grid resolution and refinement area are dynamically adjusted and optimized in real time to finely capture and identify hidden structures in real time.

7. The concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 1, characterized in that, The seismic wave signal identification module obtains the mine seismic wave propagation signals collected in real time, identifies the source location, magnitude, and propagation path characteristics of the seismic waves, and transmits the identified characteristics to the regional difference analysis module.

8. The concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 1, characterized in that, The microseismic multi-scale intelligent analysis module dynamically determines the frequency and amplitude of threshold update by calculating the sample entropy change trend of the data sequence within the sliding window in real time. The real-time sample entropy threshold is used to guide the consistent random sampling strategy. When the sample entropy exceeds the current dynamic threshold, the distance between sampling points is shortened to increase the local sampling density, and the number of sampling points in this area is increased by 20% - 30% additionally. When the sample entropy is lower than the current dynamic threshold, the distance between sampling points is extended, and the number of sampling points is reduced to 60% - 80% of the original design value, adaptively and dynamically adjusting the spatial distribution and density of sampling points.

9. An undetected structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 7, characterized in that, The regional difference analysis module obtains the rock elastic modulus, Poisson's ratio, density, and medium damping parameters of different regions of the mine and the real-time propagation data of seismic waves in real time, establishes the geological characteristic model of each region, uses the real-time propagation data of seismic waves to solve the spatial partial derivative of the seismic wave data collected at consecutive moments, obtains the wave field spatial gradient distribution, analyzes the local wave field error in real time according to the wave field spatial gradient, sets the dynamic adjustment threshold, evaluates the local Courant stability condition in real time by analyzing the local wave velocity difference and the seismic waveform frequency change trend in real time, automatically calculates the upper limit of the time step based on the real-time seismic wave velocity data, and dynamically selects the optimal time step that meets the stability requirements to accurately simulate the spatial differentiation characteristics of the seismic wave propagation path.

10. The concealed structure detection system for intelligent monitoring, early warning and prevention of mine water disasters according to claim 1, characterized in that, The intelligent data processing module monitors the change rate of input data features in real time by using the sliding window method, calculates the degree of abnormality of features by using the exponentially weighted moving average method with real-time update, adjusts the update frequency of the spatial weight matrix and the distribution density of weights in real time based on the current degree of abnormality, realizes the hierarchical interaction between local features and global features through the hierarchical attention mechanism, the local attention layer quickly responds to the location where the abnormality appears, the global attention layer dynamically coordinates the attention intensity of each local feature, and triggers the adaptive attention adjustment algorithm to quickly focus on the current most critical risk area by feeding back the data of the high-risk area of hidden structures in real time.

Images