Coal mine roof dynamic monitoring and early warning method and system based on artificial intelligence

Abstract

The application provides a coal mine roof dynamic monitoring and early warning method and system based on artificial intelligence, relates to the field of coal mine safety, and comprises the following steps: collecting a coal mine roof stress waveform and a displacement waveform, extracting a stress relaxation feature, and constructing a three-dimensional stress conduction intensity field; a stress mutation transition point is recognized by using a recurrent neural network to form a stress conduction chain; a stress field instability path is extracted by analyzing regional topological migration rules; a prediction window is determined by matching historical data; and a support reinforcement scheme is generated by continuously verifying the instability path. The application can predict the instability risk of a coal mine roof in advance, and improve the accuracy and timeliness of early warning.

Description

Technical Field

[0001] This invention relates to coal mine safety technology, and more particularly to a method and system for dynamic monitoring and early warning of coal mine roof based on artificial intelligence. Background Technology

[0002] Roof collapses caused by roof instability are one of the main threats to coal mine safety during mining operations. Dynamic monitoring of the coal mine roof is a crucial means of preventing roof collapses. Traditional roof monitoring technologies primarily rely on single-point stress or displacement monitoring, determining roof stability by measuring changes in roof stress and deformation. With the development of sensor technology, multi-point distributed monitoring has gradually become mainstream, enabling the acquisition of more comprehensive roof status information. In recent years, with the rapid development of artificial intelligence technology, applying intelligent algorithms such as deep learning to the field of coal mine safety monitoring has become a research hotspot. Through intelligent analysis of monitoring data, the risk of roof instability can be predicted more accurately.

[0003] However, existing technologies primarily focus on changes in single parameters such as stress or displacement, lacking a comprehensive analysis of the phase relationship between stress and displacement. This makes it difficult to accurately capture early signs of stress loosening within the roof, resulting in insufficient sensitivity in early warning systems. Traditional monitoring methods typically employ static threshold judgments, which cannot effectively identify stress transmission processes and abrupt changes under complex geological conditions. In particular, the stress migration patterns under the influence of complex geological structures such as faults are difficult to accurately grasp, and historical experience data is not fully utilized to construct accurate predictive models. Consequently, the practicality and relevance of early warning information are insufficient, making it difficult to provide effective technical support for support and reinforcement. Summary of the Invention

[0004] This invention provides a method and system for dynamic monitoring and early warning of coal mine roof based on artificial intelligence, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a method for dynamic monitoring and early warning of coal mine roof collapse based on artificial intelligence, comprising:

[0006] Stress waveforms and displacement waveforms of the coal mine roof were collected from multiple monitoring points. Stress loosening characteristics were extracted based on the phase decoupling relationship between the stress waveforms and displacement waveforms. A three-dimensional stress transmission intensity field was constructed by combining the monitoring points with the spatial projection of the fault.

[0007] The stress loosening characteristics are spatiotemporally mapped in a three-dimensional stress transmission intensity field to obtain a dynamic feature sequence. A recurrent neural network is used to identify stress abrupt transition points from the dynamic feature sequence. The stress abrupt transition points are then formed into stress transmission chains according to the spatiotemporal correlation coefficient by density clustering.

[0008] The topological migration patterns of regions in the stress transmission chain whose correlation coefficients exceed a preset correlation threshold are analyzed, and the motion trajectory of the regions in three-dimensional space is extracted to obtain the stress field instability path.

[0009] Geometric feature analysis of the stress field instability path is performed and morphological matching is performed with the instability path in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree.

[0010] The stress field instability path is continuously verified within the prediction window. When the stress field instability path is verified multiple times in a row and the prediction window is less than the preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path and an early warning signal is issued.

[0011] Stress loosening characteristics are extracted based on the phase decoupling relationship between the stress waveform and the displacement waveform. A three-dimensional stress transmission intensity field is constructed by combining the monitoring points and the fault spatial projection, including:

[0012] The waveform spectrum is obtained by performing Fourier transform on the stress waveform and displacement waveform;

[0013] Waveform phase features are extracted from the waveform spectrum, and the waveform phase features are decomposed into a multi-scale wavelet decomposition to obtain a feature sequence. Waveform periodic components and waveform trend components are separated from the feature sequence, and the waveform periodic components and waveform trend components are decoupled into phase to obtain stress loosening features.

[0014] The vertical distance and projection angle from the monitoring point to the fault projection plane are calculated, and a stress transmission coefficient is constructed. The stress transmission coefficient is then nonlinearly mapped to the stress loosening characteristics to construct a three-dimensional stress transmission intensity field.

[0015] Stress loosening characteristics are spatiotemporally mapped in a three-dimensional stress conduction intensity field to obtain a dynamic feature sequence. A recurrent neural network is used to identify stress abrupt transition points from this dynamic feature sequence. Density clustering is then used to form stress conduction chains from these stress abrupt transition points according to their spatiotemporal correlation coefficients.

[0016] The stress loosening characteristics are mapped in the time domain and spatial domain in the three-dimensional stress transmission intensity field. The wave component is extracted by the time domain transformation and the distribution component is obtained by the spatial domain transformation. The wave component and the distribution component are combined to construct a spatiotemporal feature matrix.

[0017] Dynamic feature sequences are extracted from the spatiotemporal feature matrix, and the dynamic feature sequences are segmented into feature segments by time windows.

[0018] A recurrent neural network is constructed, and a memory unit and a forget gate are set in the recurrent neural network. The feature fragment is input into the recurrent neural network, and the feature change pattern is extracted through the memory unit and the forget gate. The stress change transition point and transition intensity are identified from the output end.

[0019] The time interval between stress abrupt transition points is calculated to obtain the temporal correlation, and the spatial distance between the stress abrupt transition points is calculated to obtain the spatial correlation. The temporal correlation and spatial correlation are then fused to construct a spatiotemporal correlation coefficient.

[0020] Density clustering is performed on stress abrupt transition points based on spatiotemporal correlation coefficients. The density clustering results are sorted according to the transition intensity, and the sorted results are used to construct stress conduction chains according to the conduction direction.

[0021] Analyzing the topological migration patterns of regions in the stress transmission chain whose correlation coefficients exceed a preset correlation threshold, and extracting the motion trajectory of these regions in three-dimensional space, yields the stress field instability path, including:

[0022] The temporal correlation is obtained by calculating the temporal series similarity of adjacent transmission points in the stress transmission chain, and the spatial correlation is obtained by calculating the spatial distance and angle between transmission points. The temporal correlation and spatial correlation are weighted and fused in multiple dimensions to obtain the correlation coefficient. The correlation matrix of transmission points is constructed based on the correlation coefficient.

[0023] A topological graph structure is constructed based on the correlation matrix of transmission points. Transmission points with correlation coefficients exceeding a preset correlation threshold are used as topological nodes. The connection relationships between the topological nodes are extracted to construct a transmission subgraph. The topological migration rules are extracted through the structural evolution and connection changes of the transmission subgraph.

[0024] The topological migration law is mapped to a three-dimensional spatial coordinate system. A sequence of motion trajectory vectors is constructed based on the distribution position and motion speed of the topological nodes in space. The stress field instability path is extracted based on the direction and amplitude changes of the motion trajectory vector sequence.

[0025] Geometric feature analysis of the stress field instability path is performed and morphological matching is performed with the instability paths in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree, including:

[0026] Geometric feature analysis was performed on the stress field instability path, and the curvature change sequence, crack propagation direction sequence, and roof subsidence length sequence of the roof rupture line were calculated to construct the roof instability morphology curve.

[0027] Periodic subsidence features and stress concentration points are extracted from the top plate instability morphology curve to construct a top plate instability morphology feature vector and perform standardization processing.

[0028] Extract roof instability path samples from the historical working face database, calculate the morphological feature vector of the roof instability path samples, match the morphological feature vector with the standardized roof instability morphological feature vector, construct a matching degree scoring matrix, and calculate the comprehensive matching degree based on the matching degree scoring matrix.

[0029] The samples of roof instability paths were screened and sorted based on the overall matching degree, and the working face with the best overall matching degree was selected as the reference working face.

[0030] Extract the roof collapse time point and roof failure range point from the reference working face, determine the time span of the prediction window based on the roof collapse time point, determine the spatial boundary of the prediction window based on the roof failure range point, and output the final prediction window.

[0031] Within the prediction window, the stress field instability path is continuously verified. When the stress field instability path passes multiple consecutive verifications and the prediction window is less than the preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path, and an early warning signal is issued, including:

[0032] Stress field data is collected within the prediction window. Time series and spatial distribution information are extracted from the stress field data. The fluctuation amplitude curve of the time series is calculated through fluctuation analysis. The stress gradient value of the spatial distribution is calculated through regional segmentation. The fluctuation amplitude curve and stress gradient value are combined to generate a stress field instability path evaluation sequence.

[0033] The verification parameters of the fluctuation amplitude curve and stress gradient value in the stress field instability path evaluation sequence are calculated. The verification parameters are compared with the preset safety threshold for verification. The stress field instability paths that pass the verification are recorded, and the corresponding influence area and development direction are calculated to generate the stress field instability path evolution characteristics.

[0034] The support and reinforcement range is determined based on the stress field instability path evolution characteristics, and the stress distribution is calculated based on the stress field instability path evaluation sequence within the support and reinforcement range to generate a support and reinforcement scheme.

[0035] The time difference between the prediction window and the preset evacuation time is calculated. When the time difference is less than the safety response time and the number of consecutive verifications of the stress field instability path reaches the preset number, an early warning signal is generated based on the evolution characteristics of the stress field instability path and the support and reinforcement scheme.

[0036] The verification parameters for the fluctuation amplitude curve and stress gradient value in the stress field instability path evaluation sequence are calculated. These verification parameters are then compared with a preset safety threshold. The verified stress field instability paths are recorded, and the corresponding influence area and development direction are calculated. The evolution characteristics of the stress field instability path are generated, including:

[0037] The fluctuation amplitude curve and stress gradient value are extracted from the stress field instability path evaluation sequence. The fluctuation amplitude curve is subjected to time-series difference to obtain the amplitude change rate, and the stress gradient value is subjected to spatial gradient calculation to obtain the gradient change intensity.

[0038] Spatiotemporal coupling analysis of amplitude change rate and gradient change intensity is performed to extract the cross-response coefficient between fluctuation amplitude curve and stress gradient value, and verification parameters are calculated based on the cross-response coefficient.

[0039] The verification parameters are tracked continuously over a period of time using a sliding window. When the verification parameters continuously meet the preset safety threshold over a continuous period of time, the verification is considered successful, and the stress field instability path of the successful verification is recorded.

[0040] Extract the spatial distribution nodes of the verified stress field instability path, calculate the coverage boundary of the spatial distribution nodes to obtain the influence area, perform directional fitting on the spatial distribution nodes to obtain the development direction, and combine the influence area and development direction with time changes to generate stress field instability path evolution characteristics.

[0041] A second aspect of this invention provides an artificial intelligence-based dynamic monitoring and early warning system for coal mine roof, comprising:

[0042] The first unit is used to collect stress waveforms and displacement waveforms of the coal mine roof at multiple monitoring points, extract stress loosening characteristics based on the phase decoupling relationship between the stress waveforms and displacement waveforms, and construct a three-dimensional stress transmission intensity field by combining the monitoring points and fault spatial projection.

[0043] The second unit is used to perform spatiotemporal mapping of stress loosening characteristics in a three-dimensional stress transmission intensity field to obtain a dynamic feature sequence, and to use a recurrent neural network to identify stress abrupt transition points from the dynamic feature sequence. Then, through density clustering, the stress abrupt transition points are used to form stress transmission chains according to the spatiotemporal correlation coefficient.

[0044] The third unit is used to analyze the topological migration pattern of regions in the stress transmission chain whose correlation coefficient exceeds a preset correlation threshold, and to extract the motion trajectory of the region in three-dimensional space to obtain the stress field instability path.

[0045] The fourth unit is used to perform geometric feature analysis on the stress field instability path and to match the instability path with the morphology in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree.

[0046] The fifth unit is used to continuously verify the stress field instability path within the prediction window. When the stress field instability path is verified multiple times in a row and the prediction window is less than the preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path and an early warning signal is issued.

[0047] A third aspect of the present invention provides an electronic device, comprising:

[0048] processor;

[0049] Memory used to store processor-executable instructions;

[0050] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0051] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0052] In this embodiment, stress loosening characteristics are extracted by decoupling the phase of stress waveforms and displacement waveforms. A three-dimensional stress transmission intensity field is constructed by combining monitoring points and fault spatial projection, achieving accurate characterization of the stress state of the coal mine roof and overcoming the limitations of traditional monitoring methods that rely on only a single parameter. A recurrent neural network is used to identify stress abrupt transition points and form stress transmission chains through density clustering, capturing minute changes and correlations in the stress field, improving the sensitivity and accuracy of early warning. By analyzing the topological migration patterns of high-correlation-coefficient regions in the stress transmission chain, the three-dimensional spatial motion trajectory is extracted to obtain the stress field instability path, realizing the visualization and quantitative analysis of the roof's dynamic instability process. The current stress field instability path is morphologically matched with a historical working face database, and a prediction window is determined based on instability processes under similar working conditions, enabling the early warning system to possess adaptive learning capabilities and experience knowledge accumulation functions. By continuously verifying the stress field instability path within the prediction window and setting up a multi-confirmation mechanism, the false alarm rate is significantly reduced, and targeted support and reinforcement schemes can be automatically generated based on the instability path, providing sufficient preparation time for miners' safe evacuation and roof reinforcement. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the dynamic monitoring and early warning method for coal mine roof based on artificial intelligence, as described in an embodiment of the present invention.

[0054] Figure 2 This is a flowchart illustrating the verification and early warning generation process for the instability path of the stress field within the prediction window in an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0057] Figure 1 This is a flowchart illustrating an artificial intelligence-based dynamic monitoring and early warning method for coal mine roof according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0058] Stress waveforms and displacement waveforms of the coal mine roof were collected from multiple monitoring points. Stress loosening characteristics were extracted based on the phase decoupling relationship between the stress waveforms and displacement waveforms. A three-dimensional stress transmission intensity field was constructed by combining the monitoring points with the spatial projection of the fault.

[0059] The stress loosening characteristics are spatiotemporally mapped in a three-dimensional stress transmission intensity field to obtain a dynamic feature sequence. A recurrent neural network is used to identify stress abrupt transition points from the dynamic feature sequence. The stress abrupt transition points are then formed into stress transmission chains according to the spatiotemporal correlation coefficient by density clustering.

[0060] The topological migration patterns of regions in the stress transmission chain whose correlation coefficients exceed a preset correlation threshold are analyzed, and the motion trajectory of the regions in three-dimensional space is extracted to obtain the stress field instability path.

[0061] Geometric feature analysis of the stress field instability path is performed and morphological matching is performed with the instability path in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree.

[0062] The stress field instability path is continuously verified within the prediction window. When the stress field instability path is verified multiple times in a row and the prediction window is less than the preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path and an early warning signal is issued.

[0063] In one optional implementation, extracting stress loosening characteristics based on the phase decoupling relationship between the stress waveform and the displacement waveform, and constructing a three-dimensional stress transmission intensity field by combining monitoring points and fault spatial projection includes:

[0064] The waveform spectrum is obtained by performing Fourier transform on the stress waveform and displacement waveform;

[0065] Waveform phase features are extracted from the waveform spectrum, and the waveform phase features are decomposed into a multi-scale wavelet decomposition to obtain a feature sequence. Waveform periodic components and waveform trend components are separated from the feature sequence, and the waveform periodic components and waveform trend components are decoupled into phase to obtain stress loosening features.

[0066] The vertical distance and projection angle from the monitoring point to the fault projection plane are calculated, and a stress transmission coefficient is constructed. The stress transmission coefficient is then nonlinearly mapped to the stress loosening characteristics to construct a three-dimensional stress transmission intensity field.

[0067] After acquiring stress and displacement waveform data, the coal mine roof monitoring device processes these two types of waveform data to extract stress loosening characteristics. Specifically, the acquired stress and displacement waveforms are transformed to the frequency domain using Fourier transform to obtain the corresponding waveform spectra. During the Fourier transform process, the Fast Fourier Transform (FFT) algorithm is used to process the time-domain data, converting the stress and displacement waveforms at each monitoring point into a spectral representation composed of amplitude and phase, respectively. For monitoring data with a sampling rate of 200 points per second, a spectral distribution within the 0-100Hz range can be obtained through a 1024-point FFT.

[0068] Extracting waveform phase features from the waveform spectrum is the next crucial step. Waveform phase features reflect the temporal relationship between stress and displacement waves, which is significant for identifying the loosening state of coal mine roofs. The extracted waveform phase features are further processed through multi-scale wavelet decomposition. The phase features are decomposed into five levels using the Daubechies wavelet basis function, yielding wavelet coefficients at a series of scales, forming a feature sequence. For coal mine roof stress monitoring, choosing decomposition level 5 considers the frequency characteristics of stress changes in the coal mine roof, effectively capturing stress variation characteristics from high to low frequencies.

[0069] The feature sequence contains waveform information at different frequencies and time scales. From this feature sequence, the empirical mode decomposition method is used to separate the waveform periodic component and the waveform trend component. The waveform periodic component reflects the periodic variation law of the roof stress, while the waveform trend component reflects the long-term variation trend. In specific implementation, the wavelet coefficients are reconstructed according to frequency, and the wavelet coefficients with frequencies greater than 1Hz are reconstructed into periodic components, while the wavelet coefficients with frequencies less than 1Hz are reconstructed into trend components.

[0070] Phase decoupling is performed on the separated periodic and trend components of the waveform to obtain stress loosening characteristics. During phase decoupling, the phase difference between the stress and displacement waveforms in the periodic and trend components is calculated, and the stress loosening characteristics are defined based on the rate of change of this phase difference. When loosening occurs in the coal mine roof, the phase difference between the stress and displacement waves changes significantly. In actual monitoring, a phase difference change rate exceeding 0.2 radians / minute indicates potential signs of roof loosening.

[0071] To construct a three-dimensional stress transmission intensity field, it is necessary to calculate the vertical distance and projection angle from the monitoring point to the fault projection plane, thereby constructing the stress transmission coefficient. By acquiring a three-dimensional geological model of the coal mine fault, the vertical distance from each monitoring point to the nearest fault plane and its projected position on the fault plane are calculated. For the vertical distance, the Euclidean distance method is used; for the projection angle, the angle between the monitoring point's position vector and the fault plane's normal vector is calculated. In practical applications, when the vertical distance to the monitoring point is 50 meters and the projection angle is 30 degrees, the stress transmission coefficient is approximately 0.65, indicating a moderate degree of fault influence on the monitoring point.

[0072] The stress transmission coefficient reflects the degree of fault influence at the monitoring point. A three-dimensional stress transmission intensity field is constructed by nonlinearly mapping it to stress loosening characteristics. This nonlinear mapping is implemented using a radial basis function network. The stress transmission coefficient is used as a weighting factor, multiplied by the stress loosening characteristics, and then spatially interpolated using radial basis functions to generate a continuous three-dimensional stress transmission intensity field. This intensity field is represented by a three-dimensional grid with a resolution of 5 meters, covering the entire monitoring area. In a coal mine working face with a length of 150 meters and a width of 80 meters, the constructed three-dimensional stress transmission intensity field contains approximately 24,000 grid points, each with a numerical value characterizing the stress transmission intensity.

[0073] In this embodiment, by performing phase decoupling analysis on stress and displacement waveforms, the loosening characteristics of the coal mine roof can be accurately identified, improving the sensitivity and accuracy of roof instability early warning. A three-dimensional stress transmission intensity field constructed by combining the spatial relationship between monitoring points and faults enables a visual representation of the stress transmission path, providing a spatial reference framework for subsequent dynamic characteristic analysis. This significantly improves the level of safety assurance in coal mine production and reduces the risk of safety accidents caused by roof instability.

[0074] In one optional implementation, stress loosening features are spatiotemporally mapped in a three-dimensional stress conduction intensity field to obtain a dynamic feature sequence. A recurrent neural network is used to identify stress abrupt transition points from the dynamic feature sequence. Density clustering is then used to form stress conduction chains from these stress abrupt transition points according to their spatiotemporal correlation coefficients.

[0075] The stress loosening characteristics are mapped in the time domain and spatial domain in the three-dimensional stress transmission intensity field. The wave component is extracted by the time domain transformation and the distribution component is obtained by the spatial domain transformation. The wave component and the distribution component are combined to construct a spatiotemporal feature matrix.

[0076] Dynamic feature sequences are extracted from the spatiotemporal feature matrix, and the dynamic feature sequences are segmented into feature segments by time windows.

[0077] A recurrent neural network is constructed, and a memory unit and a forget gate are set in the recurrent neural network. The feature fragment is input into the recurrent neural network, and the feature change pattern is extracted through the memory unit and the forget gate. The stress change transition point and transition intensity are identified from the output end.

[0078] The time interval between stress abrupt transition points is calculated to obtain the temporal correlation, and the spatial distance between the stress abrupt transition points is calculated to obtain the spatial correlation. The temporal correlation and spatial correlation are then fused to construct a spatiotemporal correlation coefficient.

[0079] Density clustering is performed on stress abrupt transition points based on spatiotemporal correlation coefficients. The density clustering results are sorted according to the transition intensity, and the sorted results are used to construct stress conduction chains according to the conduction direction.

[0080] Stress loosening characteristics require temporal and spatial mapping within a three-dimensional stress transmission intensity field to obtain complete spatiotemporal features. During temporal mapping, a sliding window is applied to the stress loosening characteristics of each monitoring point, with a window length of 60 minutes and a sliding step of 5 minutes, to extract the trend of stress loosening characteristics changing over time at each monitoring point. Discrete wavelet transform is then performed on the data within the window to extract the fluctuation components of the stress loosening characteristics, reflecting the short-term changes in stress in the coal mine roof. During spatial mapping, Kriging interpolation is used to map the discrete stress loosening characteristics of the monitoring points onto a grid within the three-dimensional stress transmission intensity field, obtaining the spatial distribution components of the stress loosening characteristics and reflecting the spatial distribution of stress in the coal mine roof. When using a 5-meter three-dimensional grid, a typical coal mine working face area has approximately 24,000 grid points, each with stress loosening characteristic values ​​interpolated from surrounding monitoring points.

[0081] In constructing the spatiotemporal feature matrix by combining the fluctuation and distribution components, the fluctuation components are arranged chronologically to form the time dimension, and the distribution components are arranged spatially to form the spatial dimension. These two components are orthogonally combined to form the spatiotemporal feature matrix. Each element in the matrix represents a stress loosening characteristic value at a specific time and location. The matrix size is related to the number of grid points in the monitoring area and the monitoring duration. Within a 24-hour monitoring period, the generated spatiotemporal feature matrix contains 288 time points (based on a 5-minute sliding step) and 24,000 spatial points, with a total matrix size of 288 × 24,000.

[0082] When extracting dynamic feature sequences from the spatiotemporal feature matrix, principal component analysis (PCA) is used for dimensionality reduction to extract the main change patterns of the spatiotemporal feature matrix and reduce data redundancy. Principal components with a contribution rate exceeding 85% are selected as dynamic feature sequences, typically retaining the first 10-15 principal components. The dynamic feature sequences reflect the main patterns of stress changes in the coal mine roof over time and space, effectively capturing the dynamic evolution of the stress field. Subsequently, the dynamic feature sequences are segmented according to fixed time windows to obtain feature segments. The window length is set to 12 time points, or 60 minutes, with a 50% overlap between adjacent windows. These feature segments serve as input units for a recurrent neural network, used for subsequent identification of stress abrupt transition points.

[0083] The constructed recurrent neural network adopts a long short-term memory (LSTM) network structure, comprising an input layer, hidden layers, and an output layer. The number of nodes in the input layer is consistent with the dimension of the feature fragment. The hidden layer contains 64 LSM units, each with a memory unit and a forget gate. The memory unit is responsible for storing historical feature information, and the forget gate controls the degree of retention of historical information, enabling the network to learn long-term temporal dependencies. The output layer has two nodes, representing the stress mutation probability and transition strength, respectively. The network is trained using known stress mutation events from historical monitoring data, and the network parameters are optimized using the stochastic gradient descent algorithm with a learning rate of 0.01 and 1000 training iterations. After the model converges, new feature fragments are input into the trained recurrent neural network, and the output provides the stress mutation probability and transition strength at each time point. When the stress mutation probability exceeds the threshold of 0.75, the point is identified as a stress mutation transition point, and its corresponding transition strength, time, and spatial location information are recorded.

[0084] When calculating the temporal correlation between stress abrupt transition points, the time interval is calculated for each pair of transition points, and the time interval is converted into temporal correlation using an exponential decay function. The shorter the time interval, the higher the correlation. When the time interval is 30 minutes, the temporal correlation is approximately 0.7; when the time interval is 120 minutes, the temporal correlation drops to 0.3. In the spatial correlation calculation, Euclidean distance is used to measure the spatial distance between transition points, and the spatial distance is converted into spatial correlation using a Gaussian kernel function. The closer the spatial distance, the higher the correlation. When the spatial distance is 10 meters, the spatial correlation is approximately 0.9; when the spatial distance is 50 meters, the spatial correlation drops to 0.4. The fusion of temporal and spatial correlations uses a weighted average method with a weight ratio of 6:4 to obtain a comprehensive spatiotemporal correlation coefficient. The spatiotemporal correlation coefficient reflects the degree of correlation between stress abrupt transition points, and the coefficient value ranges from 0 to 1.

[0085] When performing density clustering of stress abrupt transition points based on spatiotemporal correlation coefficients, a density-based spatial clustering algorithm is employed. A neighborhood radius of 0.6 and a minimum number of points of 4 are set, and the spatiotemporal correlation coefficient is used as the distance metric for clustering. The neighborhood radius represents the threshold of the spatiotemporal correlation coefficient for grouping two transition points into the same cluster, and the minimum number of points represents the minimum number of transition points required to form a cluster. The clustering results are sorted according to the number of transition points and the average transition intensity; higher transition intensity results in a higher ranking. The sorted clustering results are then connected according to the chronological order of stress propagation to construct a stress propagation chain. The propagation direction points from earlier-occurring transition points to later-occurring transition points, forming a complete stress propagation chain structure.

[0086] In this embodiment, the precise identification and transmission chain construction of stress abrupt changes in coal mine roof are achieved through spatiotemporal mapping and deep learning methods, demonstrating significant technical effectiveness. The use of recurrent neural networks to process dynamic feature sequences effectively captures the temporal variation patterns of the stress field and identifies minute stress abrupt transition points that are difficult to detect using traditional methods. A density clustering method based on spatiotemporal correlation coefficients enables the automatic discovery of stress transmission paths, revealing the intrinsic laws governing stress transmission in coal mine roofs. The constructed stress transmission chain visually demonstrates the evolution path of the stress field, providing crucial evidence for predicting coal mine roof instability.

[0087] In one optional implementation, analyzing the topological migration patterns of regions in the stress transmission chain whose correlation coefficients exceed a preset correlation threshold, and extracting the motion trajectory of these regions in three-dimensional space to obtain the stress field instability path includes:

[0088] The temporal correlation is obtained by calculating the temporal series similarity of adjacent transmission points in the stress transmission chain, and the spatial correlation is obtained by calculating the spatial distance and angle between transmission points. The temporal correlation and spatial correlation are weighted and fused in multiple dimensions to obtain the correlation coefficient. The correlation matrix of transmission points is constructed based on the correlation coefficient.

[0089] A topological graph structure is constructed based on the correlation matrix of transmission points. Transmission points with correlation coefficients exceeding a preset correlation threshold are used as topological nodes. The connection relationships between the topological nodes are extracted to construct a transmission subgraph. The topological migration rules are extracted through the structural evolution and connection changes of the transmission subgraph.

[0090] The topological migration law is mapped to a three-dimensional spatial coordinate system. A sequence of motion trajectory vectors is constructed based on the distribution position and motion speed of the topological nodes in space. The stress field instability path is extracted based on the direction and amplitude changes of the motion trajectory vector sequence.

[0091] First, the temporal data and spatial location information of each stress transmission point in the stress transmission chain are acquired. A stress transmission chain typically consists of a series of monitoring points distributed in space, each recording stress data that changes over time. These monitoring points can be stress sensors embedded in the soil or rock mass, or stress transmission points obtained through displacement monitoring inversion.

[0092] For any two transmission points in a stress transmission chain, the stress value change sequence within their time window is extracted, and the similarity between the two sequences is calculated using a dynamic time warping algorithm to obtain the temporal correlation. The dynamic time warping algorithm solves the problem of time axis misalignment caused by different sampling frequencies by finding the optimal matching path between two time sequences. In coal mine roof monitoring, the stress value sequence length for each transmission point is 60 sampling points, and the temporal correlation calculation result between adjacent points falls between 0 and 1, with a larger value indicating a stronger correlation. In practical applications, when the stress change trends of two transmission points are basically consistent, the temporal correlation usually exceeds 0.85; while when the stress change trends are significantly different, the temporal correlation may be as low as 0.3.

[0093] Spatial correlation calculation includes two dimensions: spatial distance and angle. Spatial distance is calculated using Euclidean distance to determine the straight-line distance between transmission points; the closer the distance, the higher the spatial correlation. Angle calculation considers the positional relationship of the transmission points relative to the working face's advancing direction. The spatial relationship between transmission points is assessed by calculating the angle between the line connecting two points and the working face's advancing direction. In coal mine roof monitoring, when the distance between transmission points is 10 meters and the angle is 15 degrees, the spatial correlation is approximately 0.92; when the distance increases to 50 meters and the angle increases to 45 degrees, the spatial correlation decreases to 0.38. When performing multi-dimensional weighted fusion of temporal and spatial correlation, a weighted average method is used, with a weight of 0.6 for temporal correlation and 0.4 for spatial correlation, to calculate the correlation coefficient. The correlation coefficient reflects the comprehensive correlation between transmission points and is used to construct the transmission point correlation matrix. The transmission point correlation matrix is ​​a symmetric matrix with the same size as the number of transmission points, and the matrix elements are the correlation coefficients between the corresponding transmission points.

[0094] When constructing the topological graph structure based on the correlation matrix of transmission points, a preset correlation threshold of 0.75 is set. Transmission point pairs with correlation coefficients exceeding this threshold are considered topological nodes, and the connection relationships between these nodes are extracted to construct a transmission subgraph. The transmission subgraph is represented by an undirected weighted graph structure, where nodes represent transmission points, edges represent the connection relationships between transmission points, and the weight of the edges is the correlation coefficient. In coal mine roof monitoring, a typical transmission subgraph contains approximately 25-40 nodes, reflecting highly correlated areas in the stress transmission network. The structural evolution of the transmission subgraph is analyzed by comparing changes in the subgraph within different time windows, including changes in the number of nodes, edge connection methods, and graph density. Connection changes are described by the addition or subtraction of edges and changes in weights. During the stress transmission process in coal mine roofs, the transmission subgraph typically exhibits an evolutionary process from loose to compact, and then to decomposition, reflecting the dynamic changes in the stress field from stability to concentration, and then to release. The topological migration pattern extraction uses a graph embedding algorithm to map the transmission subgraph to a low-dimensional feature space and extract the main patterns of graph structure changes. The trajectory in the feature space reflects the evolution path of the conduction subgraph over time, and the turning point of the trajectory corresponds to the transition point of the stress field state.

[0095] When mapping topological migration patterns to a three-dimensional coordinate system, it is necessary to convert the abstract trajectories in the graph feature space into actual spatial position sequences. Based on the spatial distribution of topological nodes, the spatial center position of the transmission subgraph within each time window is calculated. This position is obtained by a weighted average of the spatial coordinates of all nodes in the subgraph, with the weight being the node's degree centrality. Degree centrality represents the importance of a node's connectivity in the graph; connecting more nodes results in higher degree centrality. By connecting the spatial center positions of different time windows, a sequence of motion trajectory vectors is formed. This sequence describes the movement path of the stress transmission center in three-dimensional space, with each vector containing start and end positions, direction, and amplitude information. In coal mine roof monitoring, a typical motion trajectory vector sequence contains approximately 20-30 vectors, spanning the entire monitoring area.

[0096] Stress field instability paths are extracted based on the changes in direction and amplitude of motion trajectory vector sequences. Directional changes are calculated using the angle between adjacent vectors, and amplitude changes are calculated using the rate of change of vector length. When the directional change exceeds 30 degrees or the rate of change of amplitude exceeds 50%, it is marked as a trajectory inflection point; these inflection points reflect significant changes in the direction or intensity of stress field transmission. Connecting all trajectory inflection points and the paths between them forms the stress field instability path. The stress field instability path is the key motion trajectory of the stress transmission center in space, reflecting the evolution of the stress field from stability to instability. In coal mine roof monitoring applications, the stress field instability path typically appears as a curve moving from the edge of the working face towards the center and then extending along a specific direction. The path length is usually between 80-120 meters and contains 5-8 key inflection points. The geometric characteristics of the instability path, such as curvature, length, and frequency of directional changes, are important indicators for assessing the risk of roof instability.

[0097] In this embodiment, by analyzing the topological migration patterns in the stress transmission chain and extracting the three-dimensional spatial motion trajectory, the precise identification of the stress field instability path is achieved, demonstrating significant technical effectiveness. This method overcomes the limitations of traditional single-point monitoring, integrating discrete transmission points into a coherent topological structure and revealing the overall evolution law of the stress field. Through multi-dimensional correlation coefficient calculation and topological graph analysis, stress transmission patterns that are difficult to detect using conventional methods are captured. The extracted stress field instability path visually demonstrates the spatial trajectory of stress transmission in the roof slab, providing a precise basis for predicting the location of roof instability.

[0098] In one optional implementation, geometric feature analysis is performed on the stress field instability path, and morphological matching is performed with the instability paths in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree, including:

[0099] Geometric feature analysis was performed on the stress field instability path, and the curvature change sequence, crack propagation direction sequence, and roof subsidence length sequence of the roof rupture line were calculated to construct the roof instability morphology curve.

[0100] Periodic subsidence features and stress concentration points are extracted from the top plate instability morphology curve to construct a top plate instability morphology feature vector and perform standardization processing.

[0101] Extract roof instability path samples from the historical working face database, calculate the morphological feature vector of the roof instability path samples, match the morphological feature vector with the standardized roof instability morphological feature vector, construct a matching degree scoring matrix, and calculate the comprehensive matching degree based on the matching degree scoring matrix.

[0102] The samples of roof instability paths were screened and sorted based on the overall matching degree, and the working face with the best overall matching degree was selected as the reference working face.

[0103] Extract the roof collapse time point and roof failure range point from the reference working face, determine the time span of the prediction window based on the roof collapse time point, determine the spatial boundary of the prediction window based on the roof failure range point, and output the final prediction window.

[0104] When performing geometric characteristic analysis on the stress field instability path, it is necessary to calculate the curvature change sequence of the roof fracture line, the crack propagation direction sequence, and the roof subsidence length sequence. The curvature change sequence is calculated by discretizing the stress field instability path in three-dimensional space, taking a sampling point every 5 meters, and calculating the curvature of the circle formed by three adjacent points to obtain the curvature sequence distributed along the path. In coal mine roof monitoring, the curvature change sequence reflects the degree of curvature of the roof fracture line; a larger curvature value indicates a sharper path turn and a higher stress concentration in the roof. The crack propagation direction sequence is obtained by calculating the angle between the line connecting two adjacent points on the path and the working face strike, recording the main directional changes in crack propagation. In the monitoring of a coal mine working face, the angle between the crack propagation direction and the working face strike ranged from 15° to 75°, with an average of 45°, indicating that the cracks mainly propagate along the dip of the ore layer. The roof subsidence length sequence is calculated based on the vertical displacement of each point on the path, reflecting the spatial distribution of roof subsidence. In a typical working face, the length of roof subsidence varies significantly at different locations along the path, ranging from a few millimeters to a few centimeters, forming a wavy distribution. These three sequences combine to form the roof instability morphology curve, comprehensively describing the geometric characteristics of roof instability.

[0105] When extracting periodic subsidence features and stress concentration points from the roof instability morphology curve, wavelet transform is used to perform time-frequency analysis on the roof subsidence length sequence to extract the periodic features of the subsidence. For a 120-meter-long instability path, 2-4 subsidence cycles can usually be identified, with cycle lengths between 20-40 meters. Stress concentration points are determined by analyzing the extreme points of the curvature change sequence; the location corresponding to the local maximum curvature is the stress concentration point. In a typical working face, there are usually 3-5 stress concentration points, distributed at the turning points of the instability path. The periodic subsidence features and stress concentration points together constitute the roof instability morphology feature vector, which typically has 10-15 feature values, including information such as cycle length, cycle amplitude, stress concentration point location, and intensity. Standardization processing normalizes each element in the feature vector to the range of 0-1, eliminating the influence of dimensions and facilitating subsequent matching calculations.

[0106] When extracting roof instability path samples from the historical working face database, working face samples with complete monitoring records are selected. The database contains roof instability path data for 100 historical working faces. For each historical sample, a morphological feature vector is calculated using the same method to construct a historical feature library. The standardized roof instability morphological feature vector of the current working face is matched with each vector in the historical feature library, and the cosine similarity method is used to calculate the similarity between feature vectors. The similarity value ranges from 0 to 1, with a higher value indicating a higher matching degree. For a certain coal mine working face, the matching degree with the most similar working face in the historical library is 0.92, indicating that the two have extremely high similarity. Based on the matching calculation results of all historical samples, a matching degree scoring matrix is ​​constructed. The matrix rows represent the features of the current working face, the columns represent historical samples, and the matrix element values ​​are the matching degrees of the corresponding features. The comprehensive matching degree is obtained by weighted summation of the scoring matrix. The weight allocation is determined according to the importance of the features; typically, the weight of periodic subsidence features is 0.5, and the weight of stress concentration points is 0.5.

[0107] When screening and ranking samples of roof instability paths based on comprehensive matching degree, historical working faces are arranged from highest to lowest comprehensive matching degree. In the ranking results, working faces with a comprehensive matching degree exceeding 0.8 are considered high-matching samples, and typically 5-10 historical working faces meet this criterion. From the high-matching samples, the working face with the highest comprehensive matching degree is selected as the reference working face. The comprehensive matching degree of the reference working face is usually above 0.85 to ensure the reliability of the prediction results. If multiple working faces have similar matching degrees, a secondary screening can be conducted, taking into account factors such as similar geological conditions and similar mining techniques, to select the most suitable reference working face.

[0108] When extracting the roof collapse time point and roof failure range point from the reference working face, historical monitoring data of the reference working face is analyzed to determine the time interval from the formation of the stress field instability path to the roof collapse. In a case study of a coal mine working face, the reference working face showed a time interval of 4.5 hours from the formation of the stress field instability path to the roof collapse. Based on the time point of the current working face's stress field instability path formation, this time interval is added to determine the prediction window's time span. The prediction window's time span is typically set to 0.8-1.2 times the historical reference time interval to account for uncertainties caused by differences in geological conditions. The roof failure range point is determined by analyzing the spatial distribution of the roof collapse in the reference working face, including the collapse's starting position, expansion direction, and final range. The roof failure range of the reference working face is mapped to the coordinate system of the current working face, and scaling and orientation adjustments are made considering differences in working face size and orientation to determine the spatial boundary of the prediction window. The spatial boundary typically includes a polygonal region covering all areas where roof collapse is expected to occur, with a certain safety margin. The final output prediction window includes two dimensions: time span and spatial boundary, providing a basis for subsequent verification and early warning.

[0109] In this embodiment, by performing geometric feature analysis on the stress field instability path and morphological matching with historical working face data, the precise determination of the roof instability prediction window is achieved, demonstrating significant technical effectiveness. This method fully utilizes historical experience data, comprehensively characterizing the morphological features of roof instability through the extraction of multi-dimensional geometric features such as curvature changes, crack propagation direction, and roof subsidence length. The morphological matching-based prediction mechanism effectively solves the problem of traditional prediction methods relying on expert experience and lacking objective evidence, making the prediction results more reliable. By selecting the historical working face with the best matching degree as a reference, the accuracy and practicality of the prediction window are ensured.

[0110] like Figure 2 As shown, Figure 2 This is a flowchart illustrating the verification and early warning generation process for the instability path of the stress field within the prediction window in an embodiment of the present invention.

[0111] In one optional implementation, the stress field instability path is continuously verified within the prediction window. When the stress field instability path passes verification multiple times consecutively and the prediction window is less than a preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path, and an early warning signal is issued, including:

[0112] Stress field data is collected within the prediction window. Time series and spatial distribution information are extracted from the stress field data. The fluctuation amplitude curve of the time series is calculated through fluctuation analysis. The stress gradient value of the spatial distribution is calculated through regional segmentation. The fluctuation amplitude curve and stress gradient value are combined to generate a stress field instability path evaluation sequence.

[0113] The verification parameters of the fluctuation amplitude curve and stress gradient value in the stress field instability path evaluation sequence are calculated. The verification parameters are compared with the preset safety threshold for verification. The stress field instability paths that pass the verification are recorded, and the corresponding influence area and development direction are calculated to generate the stress field instability path evolution characteristics.

[0114] The support and reinforcement range is determined based on the stress field instability path evolution characteristics, and the stress distribution is calculated based on the stress field instability path evaluation sequence within the support and reinforcement range to generate a support and reinforcement scheme.

[0115] The time difference between the prediction window and the preset evacuation time is calculated. When the time difference is less than the safety response time and the number of consecutive verifications of the stress field instability path reaches the preset number, an early warning signal is generated based on the evolution characteristics of the stress field instability path and the support and reinforcement scheme.

[0116] Acquiring stress field data within the prediction window is a fundamental step in continuous validation. A network of stress sensors deployed on the coal mine roof collects stress data five times per second, covering the entire prediction window. The stress field data includes stress values, displacement values, and acoustic emission signals from each monitoring point, collectively reflecting the changes in the stress state of the coal mine roof. Extracting the time series from the stress field data involves extracting the stress value changes over time at each monitoring point, resulting in a continuous stress time series. Spatial distribution information is obtained by spatially interpolating the stress values ​​at different monitoring points at the same time, generating a stress distribution map covering the entire working face.

[0117] When calculating the time series fluctuation amplitude curve through fluctuation analysis, a sliding window method was used with a window length of 10 minutes and a sliding step of 1 minute. The standard deviation of the data within each window was calculated to obtain the amplitude value of the stress fluctuation, forming the fluctuation amplitude curve during the window sliding process. The fluctuation amplitude curve reflects the trend of stress field fluctuation over time; increased fluctuation usually indicates a decrease in roof stability. For the collected stress data, the fluctuation amplitude gradually increased from an initial 0.15 MPa to 0.45 MPa, indicating a significant increase in roof stress fluctuation. When calculating the spatially distributed stress gradient value through regional segmentation, the working face was divided into 5m × 5m grid cells. The stress gradient value, i.e., the rate of stress change between adjacent grid points, was calculated for each grid cell. The stress gradient value reflects the degree of non-uniformity in the spatial distribution of stress; a larger gradient value indicates more significant stress concentration and a higher risk of roof instability. In a specific working face, the stress gradient value reached 0.08 MPa / m near the potential fracture line of the roof, which is 3-4 times that of the normal area. The combination of fluctuation amplitude curves and stress gradient values ​​generates a stress field instability path evaluation sequence, which contains stress field characteristics in both time and space dimensions, comprehensively reflecting the stress state of the roof.

[0118] The verification parameters for the fluctuation amplitude curve and stress gradient value in the stress field instability path evaluation sequence include the fluctuation amplitude growth rate, fluctuation duration, gradient value change rate, and gradient high value region area. The fluctuation amplitude growth rate is the relative change rate of fluctuation amplitude between adjacent time windows, reflecting the speed at which stress fluctuations intensify. The fluctuation duration is the length of time the fluctuation amplitude remains higher than the background value, reflecting the stability of stress fluctuations. The gradient value change rate is the rate at which the stress gradient value changes over time, reflecting the development trend of stress concentration areas. The gradient high value region area is the area where the stress gradient value exceeds a specific threshold, reflecting the range of stress concentration. These verification parameters are compared with preset safety thresholds, which are determined based on historical data statistics. The fluctuation amplitude growth rate threshold is set to 30% / hour, the fluctuation duration threshold to 45 minutes, the gradient value change rate threshold to 50% / hour, and the gradient high value region area threshold to 15% of the working surface area. When all the above parameters exceed their respective thresholds, the stress field instability path is recorded as verified. For verified instability paths, their influence area and development direction are calculated. The area of ​​influence is the size of the stress anomaly region, obtained by integrating the area of ​​the high-value region of the gradient. The development direction is the expansion direction of the high-value region of the stress gradient, obtained by connecting the centers of the high gradient values ​​at consecutive time points. The evolution characteristics of the stress field instability path include the changing trends of the expansion rate and development direction of the area of ​​influence; these characteristics form the basis for generating support and reinforcement schemes.

[0119] When determining the support reinforcement range based on the evolution characteristics of the stress field instability path, the current area affected by the instability path is used as a basis, considering the development direction and predicting the area that may expand in the future. The reinforcement range is usually set as the current area affected plus a safety buffer zone extending along the development direction, with the buffer zone width being 50% of the width of the area affected. In a specific working face, when the area affected is 200 square meters and the development direction is offset by 15 degrees along the working face advancement direction, the support reinforcement range covers an area of ​​approximately 300 square meters. The stress distribution is calculated based on the stress field instability path evaluation sequence within the support reinforcement range, and the stress concentration degree and principal stress direction of each area are analyzed. The stress distribution is obtained by spatial interpolation of the stress gradient values ​​in the evaluation sequence, forming a continuous stress distribution map. A support reinforcement scheme is generated based on the stress distribution, including three aspects: support location, support density, and support strength. The support location prioritizes the area with the highest stress gradient, the support density is proportional to the stress gradient, and the support strength is proportional to the fluctuation amplitude. For areas with stress gradients exceeding 0.07 MPa / m, the support density is increased to 1.5 times the standard density; for areas with fluctuation amplitudes exceeding 0.4 MPa, the individual support strength is increased to 300 kN. The generated support reinforcement scheme is presented in the form of a working face coordinate diagram, indicating the support parameters for different areas, facilitating on-site implementation.

[0120] The time difference between the predicted window and the preset evacuation time is a key criterion for issuing early warning signals. The preset evacuation time is the shortest time required for safe evacuation, calculated based on the number of personnel at the work site, the length of the evacuation route, and the evacuation speed; it is typically set to 30 minutes. The time difference is the difference between the end time of the predicted window and the current time, reflecting the remaining time available for response. The safe response time is the shortest time required for support reinforcement and personnel evacuation, set to 45 minutes. When the time difference is less than the safe response time, meaning the remaining time in the predicted window is insufficient to complete the safe response, and the number of consecutive verifications of the stress field instability path reaches the preset number, the system will issue an early warning signal. The preset number is typically set to 3 to avoid false alarms caused by a single abnormal data point. Early warning signals have three levels: a yellow warning indicates the need for enhanced monitoring, an orange warning indicates the need for support reinforcement, and a red warning indicates the need for immediate personnel evacuation. Early warning signals are simultaneously issued through the control center display screen, on-site audible and visual alarms, and personnel mobile devices to ensure timely information delivery to all relevant personnel.

[0121] In this embodiment, accurate early warning of coal mine roof instability is achieved by continuously verifying the stress field instability path within the prediction window. Combining dynamic monitoring with continuous verification, and employing fluctuation analysis and regional block calculations, the stress field instability state is comprehensively assessed, improving the reliability of the early warning. The support and reinforcement scheme generated based on the evolution characteristics of the stress field instability path is highly targeted and resource-efficient, avoiding the resource waste caused by blind reinforcement in traditional methods. The early warning mechanism considers the relationship between the prediction window and evacuation time, maximizing production time while ensuring safety.

[0122] In one optional implementation, verification parameters for the fluctuation amplitude curve and stress gradient value in the stress field instability path evaluation sequence are calculated. These verification parameters are then compared with a preset safety threshold. Stress field instability paths that pass verification are recorded, and the corresponding influence area and development direction are calculated. The resulting stress field instability path evolution characteristics include:

[0123] The fluctuation amplitude curve and stress gradient value are extracted from the stress field instability path evaluation sequence. The fluctuation amplitude curve is subjected to time-series difference to obtain the amplitude change rate, and the stress gradient value is subjected to spatial gradient calculation to obtain the gradient change intensity.

[0124] Spatiotemporal coupling analysis of amplitude change rate and gradient change intensity is performed to extract the cross-response coefficient between fluctuation amplitude curve and stress gradient value, and verification parameters are calculated based on the cross-response coefficient.

[0125] The verification parameters are tracked continuously over a period of time using a sliding window. When the verification parameters continuously meet the preset safety threshold over a continuous period of time, the verification is considered successful, and the stress field instability path of the successful verification is recorded.

[0126] Extract the spatial distribution nodes of the verified stress field instability path, calculate the coverage boundary of the spatial distribution nodes to obtain the influence area, perform directional fitting on the spatial distribution nodes to obtain the development direction, and combine the influence area and development direction with time changes to generate stress field instability path evolution characteristics.

[0127] First, fluctuation amplitude curves and stress gradient values ​​were extracted from the stress field instability path evaluation sequence. The fluctuation amplitude curve reflects the change in stress field fluctuation intensity over time, and its data comes from a stress sensor network deployed on the coal mine roof. The stress sensors are arranged in a 5m × 5m grid, with a sampling frequency of 5 times per second. The fluctuation amplitude curve is generated through sliding window processing. The window length is set to 10 minutes, with a step size of 1 minute, and the standard deviation of the data within each window is calculated as the fluctuation amplitude. The stress gradient value reflects the degree of spatial non-uniformity of stress distribution and is obtained by spatially differencing the stress values ​​at different locations within the working face area. In actual monitoring at the coal mine working face, the fluctuation amplitude curve shows a gradual increasing trend over time, increasing from an initial 0.12 MPa to 0.48 MPa before the warning; the stress gradient value increases significantly near the potential fracture area of ​​the roof, reaching a maximum of 0.085 MPa / m. Temporal differencing of the fluctuation amplitude curve is obtained by calculating the difference in fluctuation amplitude between adjacent time points, reflecting the rate of change of fluctuation intensity. The temporal difference method employs forward difference to calculate the increment of the fluctuation amplitude per minute, obtaining a time series of amplitude change rates. Spatial gradient calculation of the stress gradient values ​​analyzes the second-order spatial variation of the stress gradient, calculating the spatial change rate of adjacent gradient values ​​to obtain the gradient intensity distribution. The spatial gradient calculation uses the central difference method, approximating the stress gradient field within the working surface region with a second-order differential; the results reflect the drastic spatial variation of the stress gradient.

[0128] Spatiotemporal coupling analysis of amplitude change rate and gradient change intensity is a crucial step in identifying precursors to roof instability. This analysis employs a cross-correlation method to calculate the correlation between the amplitude change rate at different time points and the corresponding gradient change intensity at spatial locations. The cross-correlation calculation is implemented using a sliding window with a window length of 30 minutes and a step size of 5 minutes. For each window, the cross-correlation coefficient is calculated between the amplitude change rate sequence and the corresponding gradient change intensity sequence, yielding a cross-response coefficient reflecting the degree of coupling between the two. The cross-response coefficient ranges from -1 to 1, with positive values ​​indicating a positive correlation and higher values ​​indicating a stronger coupling. When roof instability is imminent, the cross-response coefficient typically exceeds 0.75, indicating a high degree of synchronization between stress fluctuations and spatial gradient changes. Verification parameters calculated based on the cross-response coefficient include three indicators: cross-response duration, cross-response peak value, and cross-response growth rate. The cross-response duration is the length of time the cross-response coefficient exceeds the threshold of 0.6; the cross-response peak value is the maximum value of the cross-response coefficient within the observation period; and the cross-response growth rate is the rate of change of the cross-response coefficient over time. These three indicators together constitute the verification parameters, which are used to determine the reliability of the stress field instability path.

[0129] Continuous verification is achieved by tracking verification parameters over a continuous time period using a sliding window. The sliding window length is set to 60 minutes, with a step size of 10 minutes, and the verification parameters within each window are evaluated. Preset safety thresholds include three aspects: a cross-response duration threshold of 30 minutes, a cross-response peak value threshold of 0.75, and a cross-response growth rate threshold of 0.01 / minute. Verification is considered successful when the verification parameter within a window simultaneously meets all three threshold conditions. Continuous time-period tracking requires successful verification within three consecutive windows to determine if the stress field instability path verification is successful. This multi-window continuous verification mechanism effectively reduces the probability of false alarms caused by short-term fluctuations. In the monitoring of a coal mine working face, this mechanism successfully filtered out short-term stress fluctuations caused by blasting operations, avoiding false alarms. The verified stress field instability paths are fully recorded, including the time period, spatial distribution, and parameter characteristics. The recorded content is stored in the monitoring database to provide data support for subsequent analysis and model optimization.

[0130] Extracting the spatially distributed nodes of the verified stress field instability path is fundamental to calculating the influence area and development direction. Spatially distributed nodes refer to grid points in the stress field whose cross-response coefficient exceeds a threshold of 0.7; these points constitute the spatial skeleton of the stress field instability path. The extraction of spatially distributed nodes employs a threshold segmentation method, filtering out discrete point sets that meet the conditions from the stress field. The coverage boundary of the spatially distributed nodes is calculated using a convex hull algorithm, transforming the discrete point set into a continuous polygonal region; the area of ​​this polygon is the influence area of ​​the stress field instability path. When the number of spatially distributed nodes is 32, the calculated influence area is approximately 240 square meters, covering about 15% of the working surface. Direction fitting of the spatially distributed nodes is a method to determine the development direction of the instability path; principal component analysis (PCA) is used to extract the main distribution directions of the point set. PCA calculates the covariance matrix of the spatially distributed nodes and extracts the eigenvector corresponding to its largest eigenvalue as the development direction. The development direction is represented by the angle with the working surface orientation; for example, the development direction of the instability path of a certain working surface is at a 38-degree angle to the working surface orientation. The evolution characteristics of the stress field instability path are generated by combining the area of ​​influence and the direction of development with temporal variations. These characteristics include three aspects: the expansion rate of the area of ​​influence, the stability of the direction of development, and the changing trend of the spatial distribution node density. The expansion rate of the area of ​​influence reflects the speed of the instability path's expansion, the stability of the direction of development reflects the consistency of the instability path's direction, and the changing trend of the spatial distribution node density reflects the intensification of the instability.

[0131] In this embodiment, by calculating verification parameters of the fluctuation amplitude curve and stress gradient value in the stress field instability path evaluation sequence, accurate verification of the risk of coal mine roof instability is achieved, and weak signals before roof instability are captured through cross-response coefficients. The design of multi-index verification parameters and the multi-window continuous verification mechanism significantly improve the reliability of early warning and effectively reduce the false alarm rate. The extraction of spatially distributed nodes and the application of the convex hull algorithm enable precise definition of the influence range of the stress field instability path. Based on the development direction extraction method of principal component analysis, the expansion trend of the instability path is accurately predicted, providing precise guidance for support reinforcement.

[0132] A second aspect of this invention provides an artificial intelligence-based dynamic monitoring and early warning system for coal mine roof collapse, the system comprising:

[0133] The first unit is used to collect stress waveforms and displacement waveforms of the coal mine roof at multiple monitoring points, extract stress loosening characteristics based on the phase decoupling relationship between the stress waveforms and displacement waveforms, and construct a three-dimensional stress transmission intensity field by combining the monitoring points and fault spatial projection.

[0134] The second unit is used to perform spatiotemporal mapping of stress loosening characteristics in a three-dimensional stress transmission intensity field to obtain a dynamic feature sequence, and to use a recurrent neural network to identify stress abrupt transition points from the dynamic feature sequence. Then, through density clustering, the stress abrupt transition points are used to form stress transmission chains according to the spatiotemporal correlation coefficient.

[0135] The third unit is used to analyze the topological migration pattern of regions in the stress transmission chain whose correlation coefficient exceeds a preset correlation threshold, and to extract the motion trajectory of the region in three-dimensional space to obtain the stress field instability path.

[0136] The fourth unit is used to perform geometric feature analysis on the stress field instability path and to match the instability path with the morphology in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree.

[0137] The fifth unit is used to continuously verify the stress field instability path within the prediction window. When the stress field instability path is verified multiple times in a row and the prediction window is less than the preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path and an early warning signal is issued.

[0138] A third aspect of the present invention provides an electronic device, comprising:

[0139] processor;

[0140] Memory used to store processor-executable instructions;

[0141] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0142] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0143] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic monitoring and early warning of coal mine roof based on artificial intelligence, characterized in that, include: Stress waveforms and displacement waveforms of the coal mine roof were collected from multiple monitoring points. Stress loosening characteristics were extracted based on the phase decoupling relationship between the stress waveforms and displacement waveforms. A three-dimensional stress transmission intensity field was constructed by combining the monitoring points with the spatial projection of the fault. The stress loosening characteristics are spatiotemporally mapped in a three-dimensional stress transmission intensity field to obtain a dynamic feature sequence. A recurrent neural network is used to identify stress abrupt transition points from the dynamic feature sequence. The stress abrupt transition points are then formed into stress transmission chains according to the spatiotemporal correlation coefficient by density clustering. The topological migration patterns of regions in the stress transmission chain whose correlation coefficients exceed a preset correlation threshold are analyzed, and the motion trajectory of the regions in three-dimensional space is extracted to obtain the stress field instability path. Geometric feature analysis of the stress field instability path is performed and morphological matching is performed with the instability path in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree. The stress field instability path is continuously verified within the prediction window. When the stress field instability path is verified multiple times in a row and the prediction window is less than the preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path and an early warning signal is issued.

2. The method according to claim 1, characterized in that, Stress loosening characteristics are extracted based on the phase decoupling relationship between the stress waveform and the displacement waveform. A three-dimensional stress transmission intensity field is constructed by combining the monitoring points and the fault spatial projection, including: The waveform spectrum is obtained by performing Fourier transform on the stress waveform and displacement waveform; Waveform phase features are extracted from the waveform spectrum, and the waveform phase features are decomposed into a multi-scale wavelet decomposition to obtain a feature sequence. Waveform periodic components and waveform trend components are separated from the feature sequence, and the waveform periodic components and waveform trend components are decoupled into phase to obtain stress loosening features. The vertical distance and projection angle from the monitoring point to the fault projection plane are calculated, and a stress transmission coefficient is constructed. The stress transmission coefficient is then nonlinearly mapped to the stress loosening characteristics to construct a three-dimensional stress transmission intensity field.

3. The method according to claim 1, characterized in that, Stress loosening characteristics are spatiotemporally mapped in a three-dimensional stress conduction intensity field to obtain a dynamic feature sequence. A recurrent neural network is used to identify stress abrupt transition points from this dynamic feature sequence. Density clustering is then used to form stress conduction chains from these stress abrupt transition points according to their spatiotemporal correlation coefficients. The stress loosening characteristics are mapped in the time domain and spatial domain in the three-dimensional stress transmission intensity field. The wave component is extracted by the time domain transformation and the distribution component is obtained by the spatial domain transformation. The wave component and the distribution component are combined to construct a spatiotemporal feature matrix. Dynamic feature sequences are extracted from the spatiotemporal feature matrix, and the dynamic feature sequences are segmented into feature segments by time windows. A recurrent neural network is constructed, and a memory unit and a forget gate are set in the recurrent neural network. The feature fragment is input into the recurrent neural network, and the feature change pattern is extracted through the memory unit and the forget gate. The stress change transition point and transition intensity are identified from the output end. The time interval between stress abrupt transition points is calculated to obtain the temporal correlation, and the spatial distance between the stress abrupt transition points is calculated to obtain the spatial correlation. The temporal correlation and spatial correlation are then fused to construct a spatiotemporal correlation coefficient. Density clustering is performed on stress abrupt transition points based on spatiotemporal correlation coefficients. The density clustering results are sorted according to the transition intensity, and the sorted results are used to construct stress conduction chains according to the conduction direction.

4. The method according to claim 1, characterized in that, Analyzing the topological migration patterns of regions in the stress transmission chain whose correlation coefficients exceed a preset correlation threshold, and extracting the motion trajectory of these regions in three-dimensional space, yields the stress field instability path, including: The temporal correlation is obtained by calculating the temporal series similarity of adjacent transmission points in the stress transmission chain, and the spatial correlation is obtained by calculating the spatial distance and angle between transmission points. The temporal correlation and spatial correlation are weighted and fused in multiple dimensions to obtain the correlation coefficient. The correlation matrix of transmission points is constructed based on the correlation coefficient. A topological graph structure is constructed based on the correlation matrix of transmission points. Transmission points with correlation coefficients exceeding a preset correlation threshold are used as topological nodes. The connection relationships between the topological nodes are extracted to construct a transmission subgraph. The topological migration rules are extracted through the structural evolution and connection changes of the transmission subgraph. The topological migration law is mapped to a three-dimensional spatial coordinate system. A sequence of motion trajectory vectors is constructed based on the distribution position and motion speed of the topological nodes in space. The stress field instability path is extracted based on the direction and amplitude changes of the motion trajectory vector sequence.

5. The method according to claim 1, characterized in that, Geometric feature analysis of the stress field instability path is performed and morphological matching is performed with the instability paths in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree, including: Geometric feature analysis was performed on the stress field instability path, and the curvature change sequence, crack propagation direction sequence, and roof subsidence length sequence of the roof rupture line were calculated to construct the roof instability morphology curve. Periodic subsidence features and stress concentration points are extracted from the top plate instability morphology curve to construct a top plate instability morphology feature vector and perform standardization processing. Extract roof instability path samples from the historical working face database, calculate the morphological feature vector of the roof instability path samples, match the morphological feature vector with the standardized roof instability morphological feature vector, construct a matching degree scoring matrix, and calculate the comprehensive matching degree based on the matching degree scoring matrix; The samples of roof instability paths were screened and sorted based on the overall matching degree, and the working face with the best overall matching degree was selected as the reference working face. Extract the roof collapse time point and roof failure range point from the reference working face, determine the time span of the prediction window based on the roof collapse time point, determine the spatial boundary of the prediction window based on the roof failure range point, and output the final prediction window.

6. The method according to claim 1, characterized in that, Within the prediction window, the stress field instability path is continuously verified. When the stress field instability path passes multiple consecutive verifications and the prediction window is less than the preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path, and an early warning signal is issued, including: Stress field data is collected within the prediction window. Time series and spatial distribution information are extracted from the stress field data. The fluctuation amplitude curve of the time series is calculated through fluctuation analysis. The stress gradient value of the spatial distribution is calculated through regional segmentation. The fluctuation amplitude curve and stress gradient value are combined to generate a stress field instability path evaluation sequence. The verification parameters of the fluctuation amplitude curve and stress gradient value in the stress field instability path evaluation sequence are calculated. The verification parameters are compared with the preset safety threshold for verification. The stress field instability paths that pass the verification are recorded, and the corresponding influence area and development direction are calculated to generate the stress field instability path evolution characteristics. The support and reinforcement range is determined based on the stress field instability path evolution characteristics, and the stress distribution is calculated based on the stress field instability path evaluation sequence within the support and reinforcement range to generate a support and reinforcement scheme. The time difference between the prediction window and the preset evacuation time is calculated. When the time difference is less than the safety response time and the number of consecutive verifications of the stress field instability path reaches the preset number, an early warning signal is generated based on the evolution characteristics of the stress field instability path and the support and reinforcement scheme.

7. The method according to claim 6, characterized in that, The verification parameters for the fluctuation amplitude curve and stress gradient value in the stress field instability path evaluation sequence are calculated. These verification parameters are then compared with a preset safety threshold. The verified stress field instability paths are recorded, and the corresponding influence area and development direction are calculated. The evolution characteristics of the stress field instability path are generated, including: The fluctuation amplitude curve and stress gradient value are extracted from the stress field instability path evaluation sequence. The fluctuation amplitude curve is subjected to time-series difference to obtain the amplitude change rate, and the stress gradient value is subjected to spatial gradient calculation to obtain the gradient change intensity. Spatiotemporal coupling analysis of amplitude change rate and gradient change intensity is performed to extract the cross-response coefficient between fluctuation amplitude curve and stress gradient value, and verification parameters are calculated based on the cross-response coefficient. The verification parameters are tracked continuously over a period of time using a sliding window. When the verification parameters continuously meet the preset safety threshold over a continuous period of time, the verification is considered successful, and the stress field instability path of the successful verification is recorded. Extract the spatial distribution nodes of the verified stress field instability path, calculate the coverage boundary of the spatial distribution nodes to obtain the influence area, perform directional fitting on the spatial distribution nodes to obtain the development direction, and combine the influence area and development direction with time changes to generate stress field instability path evolution characteristics.

8. An artificial intelligence-based dynamic monitoring and early warning system for coal mine roof, used to implement the method of any one of claims 1-7, characterized in that, include: The first unit is used to collect stress waveforms and displacement waveforms of the coal mine roof at multiple monitoring points, extract stress loosening characteristics based on the phase decoupling relationship between the stress waveforms and displacement waveforms, and construct a three-dimensional stress transmission intensity field by combining the monitoring points and fault spatial projection. The second unit is used to perform spatiotemporal mapping of stress loosening characteristics in a three-dimensional stress transmission intensity field to obtain a dynamic feature sequence, and to use a recurrent neural network to identify stress abrupt transition points from the dynamic feature sequence. Then, through density clustering, the stress abrupt transition points are used to form stress transmission chains according to the spatiotemporal correlation coefficient. The third unit is used to analyze the topological migration pattern of regions in the stress transmission chain whose correlation coefficient exceeds a preset correlation threshold, and to extract the motion trajectory of the region in three-dimensional space to obtain the stress field instability path. The fourth unit is used to perform geometric feature analysis on the stress field instability path and to match the instability path with the morphology in the historical working face database. The prediction window is determined based on the instability process of the working face with the best matching degree. The fifth unit is used to continuously verify the stress field instability path within the prediction window. When the stress field instability path is verified multiple times in a row and the prediction window is less than the preset evacuation time, a support and reinforcement scheme is generated based on the stress field instability path and an early warning signal is issued.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images