Mine multi-system collaborative early warning method and system based on three-dimensional GIS digital twinning

Abstract

The present application belongs to the field of early warning technology, and particularly relates to a mine multi-system collaborative early warning method and system based on three-dimensional GIS digital twinning. The method comprises: obtaining time series monitoring data of multiple monitoring systems for standardization, and mapping to a three-dimensional GIS model to construct a space-time data matrix; calculating a prior risk weight based on fault and ventilation information to generate a spatial risk weight matrix; constructing an objective function containing a data fidelity term and a mixed sparsity constraint term, calculating the norm ratio after solving, adjusting the sparsity component weight when the threshold value is exceeded, and applying a negative correlation penalty factor to the energy maximum overlap group; using the risk weight reciprocal to weight the constraint term, and using the alternating direction multiplier method to obtain a sparse solution matrix, and comparing the multi-level early warning threshold and triggering the early warning in the model. The scheme of the present application can improve the accuracy of mine multi-source monitoring data processing and the reliability of safety early warning, and enhance the accuracy and intuitiveness of mine disaster collaborative prevention and control.

Description

Technical Field

[0001] This invention belongs to the field of early warning technology, and in particular relates to a collaborative early warning method and system for multiple systems in mines based on three-dimensional GIS digital twins. Background Technology

[0002] Modern mines typically deploy sensors covering multiple independent monitoring systems, including those for microseismic activity, gas monitoring, ventilation, and stress monitoring. The development and occurrence of mine disasters are often closely related to the three-dimensional spatial topology and geological structure underground. Traditional methods cannot deeply integrate massive amounts of real-time monitoring data with key spatial entity features such as fault structures, goaf areas, and ventilation paths. This early warning model, which severs the connection between temporal monitoring and the real three-dimensional environment, results in the system lacking control over the a priori characteristics of overall mine spatial risks, and cannot achieve truly intuitive multi-system collaborative early warning. The deep monitoring environment in mines is highly uncertain and has a high noise background. Abnormal precursor signals generated in the early stages of disaster development are often extremely weak and exhibit complex overlapping or isolated characteristics in their spatiotemporal distribution. Existing early warning signal extraction and anomaly detection algorithms generally suffer from poor model adaptability and weak tracking capabilities when processing such high-dimensional and complex spatiotemporal data. Traditional signal processing models mostly employ fixed mathematical constraints, failing to adjust for the evolution of anomalous signal energy across different monitoring periods, and unable to embed spatially differentiated geological prior risk weights into the anomalous feature extraction process. This signal analysis method, lacking spatial prior guidance, fails to detect and isolate the weak signals truly representing disaster anomalies under strong noise interference. Because it cannot handle variations in anomalous signals between clusters and isolated points, existing early warning systems often face high false alarm and false negative rates, limiting the accuracy and reliability of collaborative early warning systems in mines under complex geological conditions. Summary of the Invention

[0003] To achieve accurate and coordinated early warning across multiple systems, this disclosure provides a method and system for coordinated early warning of multiple systems in mines based on three-dimensional GIS digital twins.

[0004] In a first aspect, embodiments of this disclosure provide a collaborative early warning method for multiple systems in a mine based on three-dimensional GIS digital twins, comprising the following steps: The time-series monitoring data of the mine's multi-monitoring system is acquired and standardized, and mapped with spatial entities in the 3D GIS model to construct a spatiotemporal data matrix. Based on the pre-set fault structure, goaf, and ventilation path information in the 3D GIS model, the prior risk weight of each spatial location is calculated to generate a spatial risk weight matrix. A warning solution objective function is constructed, which includes a data fidelity term and a mixed sparsity constraint term. The mixed sparsity constraint term is composed of a weighted sum of a first sparsity component and a second sparsity component. The weight coefficients between the components are initialized. After solving the current warning cycle, the ratio of the non-zero norm contributed by each overlapping group to the non-zero norm contributed by isolated elements in the sparse solution matrix is ​​calculated. When the ratio exceeds a predetermined threshold, the weight of the second sparsity component in the next cycle is increased. For the specific overlapping group with the largest non-zero element energy in the current cycle, a penalty factor negatively correlated with the energy value is applied in the next cycle. The mixed sparsity constraint term is weighted using the reciprocal of each element in the spatial risk weight matrix, and the objective function is solved iteratively using the alternating direction multiplier method to obtain a sparse solution matrix representing the abnormal signal. The non-zero element values ​​in the sparse solution matrix are compared with multi-level early warning thresholds, and if they exceed the thresholds, the corresponding level of early warning is triggered in the three-dimensional GIS model.

[0005] Optionally, the step of acquiring time-series monitoring data from the mine's multi-monitoring system, standardizing it, and mapping it to spatial entities within a 3D GIS model to construct a spatiotemporal data matrix includes: Parse the raw messages sent from each monitoring system to extract the unique sensor identifiers and measurement values; Query the preset spatial configuration database and obtain the three-dimensional spatial absolute coordinates of the corresponding sensor in the three-dimensional GIS model based on the unique identifier; An initial two-dimensional spatiotemporal grid is constructed using the total number of sensors as the column dimension of the matrix and the sampling time series within a preset time window as the row dimension of the matrix. After the measurement values ​​are standardized, they are filled into the corresponding nodes of the initial two-dimensional spatiotemporal grid according to the corresponding timestamp and the column index corresponding to the unique identifier, thereby generating the spatiotemporal data matrix.

[0006] Optionally, the step of calculating the prior risk weights for each spatial location based on the pre-defined fault structure, goaf, and ventilation path information in the 3D GIS model, and generating a spatial risk weight matrix, includes: Calculate the shortest geometric distance from the spatial entity to the nearest fault rupture surface, and calculate the first risk weighting factor based on the ratio of the preset hazard impact radius to the distance; Determine whether the spatial entity is located inside the closed envelope of the goaf. If it is inside, assign a preset constant greater than zero to the second risk weight factor; if it is outside, assign a value of zero. Obtain the absolute wind pressure difference between the actual monitored wind pressure and the standard wind pressure at the ventilation node where the spatial entity is located, and normalize the wind pressure difference as the third risk weighting factor. The first risk weight factor, the second risk weight factor, and the third risk weight factor are multiplied by their respective preset fixed weight coefficients and summed to obtain the prior risk weights for each spatial location, thus generating the spatial risk weight matrix.

[0007] Optionally, the construction of the early warning solution objective function includes a data fidelity term and a mixed sparsity constraint term, wherein the mixed sparsity constraint term is a weighted sum of a first sparsity component and a second sparsity component, including: Subtract the preset normal security background reference matrix from the spatiotemporal data matrix to obtain the differential spatiotemporal matrix. The data fidelity term is constructed using the square of the Frobenius norm of the error between the differential spatiotemporal matrix and the sparse solution matrix to be solved. The sparse solution matrix to be solved is divided into multiple overlapping sub-matrix groups. The L2 norm of each sub-matrix group is calculated. The first sparsity component is constructed by the sum of the L2 norms of all sub-matrix groups to constrain the spatially continuous overlapping group anomalous signals. The L1 norm is calculated for each element in the sparse solution matrix to be solved, and the second sparsity component is constructed by the sum of the L1 norms of all elements to constrain isolated anomalous signals.

[0008] Optionally, the step of calculating the ratio of the norm of non-zero elements contributed by each overlapping group to the norm of non-zero elements contributed by isolated elements in the sparse solution matrix after solving the current early warning cycle includes: Extract the sparse solution matrix obtained in the current early warning period; Calculate the sum of squares of all non-zero elements in each overlapping group of the sparse solution matrix, and sum the calculation results of all overlapping groups to obtain the sum of the norms of the non-zero elements in the overlapping groups; Find the union of all matrix nodes covered by overlapping groups, traverse the independent non-zero elements in the sparse solution matrix that are not covered by the union, calculate the sum of squares of the independent non-zero elements, and obtain the sum of norms of isolated non-zero elements. The ratio used to adjust the weighting coefficients is obtained by dividing the sum of the norms of the non-zero elements in the overlapping group by the sum of the norms of the non-zero elements in the isolated group.

[0009] Optionally, applying a penalty factor negatively correlated with the energy value to the specific overlapping group with the highest non-zero element energy in the current period in the next period includes: Calculate the sum of squares of all elements in each overlapping group in the current periodic sparse solution matrix, determine the overlapping group with the largest sum of squares as the specific overlapping group, and use the sum of squares as the energy value; A penalty factor decay function is constructed, wherein the output value of the penalty factor decay function decreases exponentially with the increase of the input energy value, and the range of the output value is limited to the interval greater than 0 and not greater than 1. The energy value of the specific overlapping group is input into the penalty factor decay function to calculate the penalty factor. When constructing the objective function for the next early warning period, the corresponding term of the specific overlapping group in the first sparsity component is multiplied by the penalty factor.

[0010] Optionally, comparing the non-zero element values ​​in the sparse solution matrix with multi-level early warning thresholds, and triggering a corresponding level of early warning in the 3D GIS model when the thresholds are exceeded, includes: Extract the non-zero elements from the sparse solution matrix, and divide the absolute values ​​of the elements into multiple anomaly levels according to a preset numerical interval mapping rule; Based on the row and column indices of the elements in the sparse solution matrix, the timestamp of the anomaly and its spatial coordinates in the 3D GIS model are obtained by reverse parsing. At the corresponding spatial coordinates of the 3D GIS model, a 3D highlighted marker with a color corresponding to the hazard level is generated, and the data panel of the associated monitoring system topologically connected to the spatial location is automatically retrieved and displayed in a pop-up window.

[0011] Secondly, this disclosure also provides a multi-system collaborative early warning system for mines based on three-dimensional GIS digital twins, including the following modules: The generation module is used to acquire time-series monitoring data from multiple monitoring systems in the mine, perform standardized processing, and map it with spatial entities in the 3D GIS model to construct a spatiotemporal data matrix. Based on the pre-set fault structure, goaf, and ventilation path information in the 3D GIS model, the module calculates the prior risk weights of each spatial location and generates a spatial risk weight matrix. The calculation module is used to construct an early warning solution objective function that includes a data fidelity term and a mixed sparsity constraint term. The mixed sparsity constraint term is composed of a weighted sum of a first sparsity component and a second sparsity component. The weight coefficients between the components are initialized. After solving the problem in the current early warning cycle, the ratio of the non-zero norm contributed by each overlapping group to the non-zero norm contributed by isolated elements in the sparse solution matrix is ​​calculated. When the ratio exceeds a predetermined threshold, the weight of the second sparsity component in the next cycle is increased. For the specific overlapping group with the largest non-zero element energy in the current cycle, a penalty factor negatively correlated with the energy value is applied in the next cycle. The early warning module is used to weight the mixed sparsity constraint terms by using the reciprocals of each element in the spatial risk weight matrix, and to iteratively solve the objective function using the alternating direction multiplier method to obtain a sparse solution matrix representing the abnormal signal. The non-zero element values ​​in the sparse solution matrix are compared with multi-level early warning thresholds, and if they exceed the thresholds, the corresponding level of early warning is triggered in the three-dimensional GIS model.

[0012] Preferably, the step of acquiring time-series monitoring data from the mine's multi-monitoring system, standardizing the data, and mapping it to spatial entities within a 3D GIS model to construct a spatiotemporal data matrix includes: Parse the raw messages sent from each monitoring system to extract the unique sensor identifiers and measurement values; Query the preset spatial configuration database and obtain the three-dimensional spatial absolute coordinates of the corresponding sensor in the three-dimensional GIS model based on the unique identifier; An initial two-dimensional spatiotemporal grid is constructed using the total number of sensors as the column dimension of the matrix and the sampling time series within a preset time window as the row dimension of the matrix. After the measurement values ​​are standardized, they are filled into the corresponding nodes of the initial two-dimensional spatiotemporal grid according to the corresponding timestamp and the column index corresponding to the unique identifier, thereby generating the spatiotemporal data matrix.

[0013] Preferably, the step of calculating the prior risk weights for each spatial location and generating a spatial risk weight matrix based on the pre-set fault structure, goaf, and ventilation path information in the 3D GIS model includes: Calculate the shortest geometric distance from the spatial entity to the nearest fault rupture surface, and calculate the first risk weighting factor based on the ratio of the preset hazard impact radius to the distance; Determine whether the spatial entity is located inside the closed envelope of the goaf. If it is inside, assign a preset constant greater than zero to the second risk weight factor; if it is outside, assign a value of zero. Obtain the absolute wind pressure difference between the actual monitored wind pressure and the standard wind pressure at the ventilation node where the spatial entity is located, and normalize the wind pressure difference as the third risk weighting factor. The first risk weight factor, the second risk weight factor, and the third risk weight factor are multiplied by their respective preset fixed weight coefficients and summed to obtain the prior risk weights for each spatial location, thus generating the spatial risk weight matrix.

[0014] The beneficial effects of this invention are as follows: This invention improves the accuracy of multi-source monitoring data processing and the reliability of safety early warning in mines. By mapping the temporal monitoring data with the spatial entities in a 3D GIS model and combining it with key geological and production environment information such as fault structures and goaf areas to generate a spatial risk weight matrix, the organic integration of multi-source spatiotemporal data and the assessment of prior risks are achieved. When constructing the early warning objective function, a mixed sparsity constraint is input, and the sparsity component weights are flexibly adjusted using the norm ratio between overlapping groups and isolated element features. Based on the energy value mechanism, a penalty factor negatively correlated with the energy value is applied to specific overlapping groups to extract potential weak anomaly signals. The spatial risk weight guides the iterative solution of the objective function, enhancing the sensitivity of high-risk areas to anomaly signals. Combined with multi-level thresholds, early warnings are triggered in a 3D visualization platform, enhancing the accuracy and intuitiveness of collaborative prevention and control of mine disasters. Attached Figure Description

[0015] Figure 1 This is a flowchart of a multi-system collaborative early warning method for mines based on 3D GIS digital twins, as described in this embodiment of the invention. Figure 2 This is a schematic diagram of the spatiotemporal data matrix in an embodiment of the present invention; Figure 3 This is a schematic diagram comparing the performance of different models in an embodiment of the present invention. Detailed Implementation

[0016] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0017] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0018] One or more embodiments of the present invention provide a collaborative early warning method for multiple systems in mines based on three-dimensional GIS digital twins, such as... Figure 1 As shown, it includes the following steps: S1. Acquire time-series monitoring data from the mine's multi-monitoring system and perform standardized processing. Map the data to spatial entities within the 3D GIS model to construct a spatiotemporal data matrix. Based on the pre-set fault structures, goaf areas, and ventilation path information in the 3D GIS model, calculate the prior risk weights for each spatial location and generate a spatial risk weight matrix.

[0019] Real-time data streams from microseismic monitoring sensors, gas concentration sensors, and stress sensors are subscribed to through the Apache Kafka distributed event stream processing platform. The `fit_transform` algorithm function of the `MinMaxScaler` class is used to perform Min-Max normalization to the [0,1] interval for multi-source heterogeneous time-series monitoring data, eliminating dimensional differences between different parameters. Based on the PostGIS spatial relation database, the `ST_Contains` and `ST_Intersects` spatial topology query functions are executed to accurately bind the standardized 3D geographic coordinates of sensor nodes to the topological node entities of the underground tunnel model rendered by the Cesium open-source 3D Earth engine. A two-dimensional spatiotemporal data matrix is ​​initialized and filled, with columns representing 3D spatial location numbers and rows representing discrete time steps. Using the NetworkX complex network analysis graph computation library, a directed graph data structure for the entire mine ventilation network is constructed. The `dijkstra_path_length` shortest path algorithm is called to calculate the shortest spatial topological connectivity distance from each sensor monitoring point to the nearest fault structure and the polygonal boundary of the goaf. An inverse distance weighted spatial interpolation algorithm (IDW) is used to calculate the spatial distance attenuation factor for each monitoring point. This attenuation factor is then multiplied by a floating-point constant representing the static hazard level assigned by a domain expert, generating prior risk weight values ​​for each discrete spatial grid. All weight values ​​are then encapsulated and output as a two-dimensional spatial risk weight matrix with row and column indices.

[0020] In some embodiments, the step of acquiring time-series monitoring data from a multi-monitoring system in a mine, performing standardization processing, and mapping it to spatial entities within a 3D GIS model to construct a spatiotemporal data matrix includes: Parse the raw messages sent from each monitoring system to extract the unique sensor identifiers and measurement values; Query the preset spatial configuration database and obtain the three-dimensional spatial absolute coordinates of the corresponding sensor in the three-dimensional GIS model based on the unique identifier; An initial two-dimensional spatiotemporal grid is constructed using the total number of sensors as the column dimension of the matrix and the sampling time series within a preset time window as the row dimension of the matrix. After the measurement values ​​are standardized, they are filled into the corresponding nodes of the initial two-dimensional spatiotemporal grid according to the corresponding timestamp and the column index corresponding to the unique identifier, thereby generating the spatiotemporal data matrix.

[0021] The system parses raw JSON or XML messages from monitoring systems such as gas, microseismic, and roof pressure. Using regular expressions or preset key-value pair matching, it extracts a 48-bit MAC address or a 128-bit UUID as the sensor's unique identifier and extracts floating-point measurement values, such as gas concentration 0.45% and microseismic energy 1200J. Connecting to a MySQL or PostgreSQL spatial configuration database, it executes an SQL query to obtain the sensor's absolute Gauss-Kruger projection coordinates (X, Y, Z) in the mine's 3D GIS model based on the UUID match. For example, X=3854321.45, Y=543210.12, Z=-650.50m.

[0022] The column dimension is defined by N sensors deployed within the mine, preferably ranging from 500 to 2000. The row dimension is defined by a sampling frequency within a preset time window, such as T = 10 minutes, for example, 1 Hz, resulting in a total of M = 600 sampling points. A two-dimensional spatiotemporal grid empty matrix of size M × N is initialized in memory. After processing the measured values ​​to the [0,1] interval using the Min-Max normalization method, the row index is calculated based on the UNIX timestamps inherent in the measured values, and the column index is determined based on the UUID mapping table. The normalized values, for example, 0.68, are filled into the corresponding coordinate nodes of the initial two-dimensional spatiotemporal grid. For individual missing data due to network latency or equipment failure, cubic spline interpolation or kriging interpolation algorithms are used to fill in the spatiotemporal dimension, generating a structured, dense, and dimensionless spatiotemporal data matrix.

[0023] In some embodiments, the step of calculating the prior risk weights for each spatial location and generating a spatial risk weight matrix based on the pre-defined fault structures, goaf areas, and ventilation path information in the 3D GIS model includes: Calculate the shortest geometric distance from the spatial entity to the nearest fault rupture surface, and calculate the first risk weighting factor based on the ratio of the preset hazard impact radius to the distance; Determine whether the spatial entity is located inside the closed envelope of the goaf. If it is inside, assign a preset constant greater than zero to the second risk weight factor; if it is outside, assign a value of zero. Obtain the absolute wind pressure difference between the actual monitored wind pressure and the standard wind pressure at the ventilation node where the spatial entity is located, and normalize the wind pressure difference as the third risk weighting factor. The first risk weight factor, the second risk weight factor, and the third risk weight factor are multiplied by their respective preset fixed weight coefficients and summed to obtain the prior risk weights for each spatial location, thus generating the spatial risk weight matrix.

[0024] The algorithm uses a 3D spatial analysis method to calculate spatial entities such as sensor coordinates, and then calculates the 3D Euclidean shortest distance from these points to a fault fracture surface pre-constructed in a 3D GIS using geological exploration point cloud data. This distance is fitted by a large number of triangular mesh patches. Set the radius of influence of fault hazard. The preferred range is 50 meters to 150 meters, for example. =100 meters, when the distance Less than At that time, the first risk weighting factor is calculated using the linear decay formula. ,For example =40 meters =0.6; when Greater than or equal to At that time, once it was determined that they were out of danger, they were ordered to... =0. The three-dimensional coordinates of the spatial entity are determined using the three-dimensional ray cross method to determine whether they are contained within the polyhedron model of the closed envelope surface of the goaf. If they are, a Boolean truth value is triggered, and the second risk weight factor is adjusted. The value is assigned as a preset constant representing a high-risk state, with a preferred range of 0.6-1.0. For example, a value of 0.8 is assigned, while a value of 0 is assigned if the risk is outside the range.

[0025] The real-time Pitot tube differential pressure transmitter monitoring wind pressure (e.g., 245 Pa) of the ventilation node corresponding to the roadway branch where the spatial entity is located is obtained by querying the topology network. The theoretical standard wind pressure (e.g., 260 Pa) output by the ventilation network solution model is also retrieved to calculate the absolute wind pressure difference. The third risk weighting factor is obtained by using a truncated normalization method based on a preset maximum permissible wind pressure difference deviation threshold, such as 50 Pa. For example, the current =0.3. Assign fixed weight coefficients representing relative importance to the three factors mentioned above. Preferably, the first, second, and third weight coefficients are set to 0.4, 0.4, and 0.2 respectively, ensuring the total weight sum is 1. This is achieved through linear combination and by adding a preset, minimal positive constant to prevent division by zero, such as... Calculate the prior risk weight of a single point location. The result is 0.30001. After traversing and covering all N spatial nodes, the generated one-dimensional static spatial risk weight array is fully copied and filled along the M sampling time points of the preset time window to generate a spatial risk weight matrix consistent with the original data dimension M×N, as shown below. Figure 2 As shown, this is the visualization result of the spatiotemporal data matrix after normalization and filling. The horizontal axis represents the sensor number, and the vertical axis represents the monitoring duration. The locally highlighted areas in the figure represent preset continuous abnormal signal areas, and the dark background represents the normal environmental noise of the mine.

[0026] S2, construct an early warning solution objective function that includes a data fidelity term and a mixed sparsity constraint term. The mixed sparsity constraint term is composed of a weighted sum of the first sparsity component and the second sparsity component. Initialize the weight coefficients between the components. After solving in the current early warning cycle, calculate the ratio of the non-zero norm contributed by each overlapping group to the non-zero norm contributed by isolated elements in the sparse solution matrix. When the ratio exceeds a predetermined threshold, increase the weight of the second sparsity component in the next cycle. For the specific overlapping group with the largest non-zero element energy in the current cycle, apply a penalty factor negatively correlated with the energy value in the next cycle.

[0027] A mathematical programming objective function is constructed, consisting of a least-squares data fidelity term calculated based on the matrix Frobenius norm and a mixed sparsity constraint term based on the linear addition of the mixed L1 norm and the L21 norm of the overlapping group. The first sparsity component is set as the L21 norm of the overlapping group, used to constrain and extract spatially topologically continuous regional cluster anomalies. The second sparsity component is set as the standard L1 norm, used to extract isolated random sensor anomalies that have not spatially clustered. Two single-precision floating-point variables are declared in the Python code as initial weight coefficients for the two components, and constraint equations are written to ensure that their sum is always equal to 1. After solving for the sparse solution matrix representing the anomalies using an optimization algorithm within the current time window, the `linalg.norm` linear algebraic norm calculation function is used to calculate the sum of the squares of the L2 norms of the matrix subset slices belonging to each preset spatial overlapping group, and the sum of the squares of the independent matrix elements not belonging to any overlapping group. The ratio of the sum of the squares of the non-zero elements of the overlapping group to the sum of the squares of the non-zero elements of the isolated group is calculated.

[0028] Calculate the ratio of the sum of squares of non-zero elements in the overlapping group to the sum of squares of non-zero elements in the isolated group. When this ratio is greater than the empirical floating-point threshold constant hard-coded in the code, accumulate the weight variable of the second sparse component in the next cycle's control equation by a fixed incrementing step size, and subtract the same step size from the weight variable of the first sparse component. Call the NumPy library's argmax maximum index function to locate the index of the specific spatial overlapping group with the largest calculated L2 norm scalar value in the current sparse solution matrix. After extracting the maximum energy value of this group, substitute the energy value into the exponential decay function with the natural constant e as the base, and obtain a penalty factor between 0 and 1 by calculating the negative energy value of e. In the code for constructing the objective function of the next time warning cycle, multiply this penalty factor as a multiplier with the original regularization coefficient of this specific overlapping group to complete the decay update of the local weights.

[0029] In some embodiments, the construction of the early warning solution objective function, which includes a data fidelity term and a mixed sparsity constraint term, wherein the mixed sparsity constraint term is a weighted sum of a first sparsity component and a second sparsity component, includes: Subtract the preset normal security background reference matrix from the spatiotemporal data matrix to obtain the differential spatiotemporal matrix. The data fidelity term is constructed using the square of the Frobenius norm of the error between the differential spatiotemporal matrix and the sparse solution matrix to be solved. The sparse solution matrix to be solved is divided into multiple overlapping sub-matrix groups. The L2 norm of each sub-matrix group is calculated. The first sparsity component is constructed by the sum of the L2 norms of all sub-matrix groups to constrain the spatially continuous overlapping group anomalous signals. The L1 norm is calculated for each element in the sparse solution matrix to be determined, and the second sparsity component is constructed by the sum of the L1 norms of all elements to constrain isolated anomalous signals.

[0030] From historical safe operating cycles, such as the past 30 consecutive days without any abnormal alarms, mean data with the same temporal characteristics are extracted to construct a normal safety background benchmark matrix of the same dimension as the real-time data. This benchmark matrix is ​​then subtracted element-wise from the currently generated M×N dimensional real-time spatiotemporal data matrix to remove periodicity and conventional background trends, generating a difference spatiotemporal matrix. A sparse solution matrix representing the components of sudden abnormal signals is constructed. The error matrix between the difference spatiotemporal matrix and the sparse solution matrix is ​​calculated, and the squared Frobenius norm of this error matrix is ​​used as a data fidelity term in the optimization objective function. This ensures that the abnormal signals extracted by the model can numerically approximate the fluctuation residuals of the actual original observations, avoiding overfitting or signal distortion. For the mixed sparsity constraint term, the L2 norms of all K sub-matrix groups are summed to form the first sparsity component in the form of a lasso. In this way, the smooth correlation of elements within the group and the overall sparsity of the structure between groups are used to constrain and extract the group of overlapping anomalous signals that are persistent in time series and continuous in space caused by the diffusion of disaster sources such as gas outbursts or surrounding rock deformation.

[0031] Based on the three-dimensional topological connections and signal propagation characteristics of mine roadways, and assuming a spatial distance threshold of 30 meters or that sensors within the same ventilation circuit are grouped together, the sparse solution matrix is ​​divided horizontally and vertically into K overlapping sub-matrix groups with shared row and column data. The square root of the sum of the squares of all elements within each sub-matrix group yields the local L2 norm. The sum of the absolute values ​​of each independent element in the sparse solution matrix is ​​then used as the second sparsity component. Leveraging the truncation and contraction properties of L1 regularization in convex optimization, randomly distributed minute environmental noise or isolated abnormal interference caused by occasional sensor drift is forced to attenuate to zero.

[0032] In some embodiments, calculating the ratio of the non-zero norm contributed by each overlapping group to the non-zero norm contributed by isolated elements in the sparse solution matrix after solving the current early warning cycle includes: Extract the sparse solution matrix obtained in the current early warning period; Calculate the sum of squares of all non-zero elements in each overlapping group of the sparse solution matrix, and sum the calculation results of all overlapping groups to obtain the sum of the norms of the non-zero elements in the overlapping groups; Find the union of all matrix nodes covered by overlapping groups, traverse the independent non-zero elements in the sparse solution matrix that are not covered by the union, calculate the sum of squares of the independent non-zero elements, and obtain the sum of norms of isolated non-zero elements. The ratio used to adjust the weighting coefficients is obtained by dividing the sum of the norms of the non-zero elements in the overlapping group by the sum of the norms of the non-zero elements in the isolated group.

[0033] After the alternating direction multiplier method completes the current t-th warning cycle (e.g., an iteration spanning 10 minutes) and converges and exits the optimization loop, the updated full sparse solution matrix is ​​extracted from the memory buffer. Floating-point variables are initialized as overlap group accumulators and isolated element accumulators. A double-loop iterates through K pre-defined overlap groups based on the mine's spatial topology. Each overlap group preferably contains 3-8 time-series nodes from adjacent sensors. Within each overlap group, non-zero elements with absolute values ​​greater than the tolerance (e.g., 1e-5) are extracted using a logical mask. The local sum of their squared values ​​is calculated, and the results from each group are aggregated into the global accumulator to obtain the sum of the norms of the non-zero elements in the overlap groups. For example, if the system identifies a clustered gas outburst event where three overlap groups contribute 15.2, 18.5, and 11.9 respectively, the sum of the overlap group norms is 45.6.

[0034] The union of the sets of matrix node positions covered by all K overlapping groups is used to construct a global index mask matrix. All matrix node positions belonging to any overlapping group are marked and excluded. For matrix elements in the overlapping intersection region, multiple overlapping groups are allowed to be repeatedly counted when calculating the energy of each local overlapping group, but they are only marked once when generating the global mask. The remaining independent non-zero elements not covered by any overlapping group are explicitly defined as isolated elements. The squares of each element are calculated and scalar summed to obtain the sum of the non-zero norms of isolated elements representing occasional noise or extremely weak independent events, for example, a cumulative value of 12.4. A floating-point division operation is performed, dividing the sum of the non-zero norms of the overlapping groups by the sum of the non-zero norms of the isolated elements to obtain a dimensionless ratio representing the degree of spatial anomaly aggregation in the mine, such as 3.68. This ratio will be passed to the next control flow as a core trigger condition. When the ratio is determined to be higher than the set empirical threshold, such as 3.0, it is determined that the current mine status presents a group disaster outbreak trend. The weight allocation in the next iteration cycle will be adjusted, and the local regularization penalty intensity will be reduced for high-risk core areas to protect the release of group disaster signals.

[0035] In some embodiments, applying a penalty factor negatively correlated with the energy value to the specific overlapping group with the highest energy of non-zero elements in the current period in the next period includes: Calculate the sum of squares of all elements in each overlapping group in the current periodic sparse solution matrix, determine the overlapping group with the largest sum of squares as the specific overlapping group, and use the sum of squares as the energy value; A penalty factor decay function is constructed, wherein the output value of the penalty factor decay function decreases exponentially with the increase of the input energy value, and the range of the output value is limited to the interval greater than 0 and not greater than 1. The energy value of the specific overlapping group is input into the penalty factor decay function to calculate the penalty factor. When constructing the objective function for the next early warning period, the corresponding term of the specific overlapping group in the first sparsity component is multiplied by the penalty factor.

[0036] The microprocessor performs matrix slicing on the sparse solution matrix obtained in the current warning period, and traverses in parallel K preset overlapping sub-matrix groups. For each overlapping group, the dot product and square operation is performed on all spatiotemporal node values ​​contained within it, and a summation operator is called to perform full summation, thereby obtaining the energy value representing the intensity of the abnormal signal in that local area. By comparing the energy values ​​of all overlapping groups through a sorting algorithm, the group with the global maximum energy peak is selected and locked as a specific overlapping group. The energy value of this specific overlapping group, such as the extracted maximum sum of squares of 85.4, maps to the core area of ​​high-risk accumulation with the most active geological dynamics or the largest catastrophic dynamic energy in the current mining area. To prevent conventional global regularization from excessively suppressing high-risk signals, a natural constant is constructed and invoked. A negative exponential penalty factor decay function with base 0.

[0037] Setting hyperparameters to control the decay rate The preferred range is 0.01 to 0.05, for example, setting... =0.02, this mathematical structure ensures that when the input energy value is greater than or equal to 0, the calculated output value will be smoothly truncated within the interval (0,1]. Substituting the selected energy value of a specific overlapping group, such as 85.4, into the exponential decay function, a penalty factor with negative correlation characteristics, such as 0.181, is calculated. When entering the next time-dimension warning cycle and re-initializing the objective function structure of the optimization algorithm, for this specific overlapping group, the fixed regularization weight coefficient originally associated with the overlapping group in the first sparse component is multiplied by the calculated penalty factor. This reduces the penalty intensity for the high-risk core area during near-end mapping solution, protecting the complete release of the contiguous disaster signal.

[0038] S3, the reciprocal of each element in the spatial risk weight matrix is ​​used to weight the mixed sparsity constraint term, and the objective function is solved iteratively using the alternating direction multiplier method to obtain a sparse solution matrix representing the abnormal signal. The non-zero element values ​​in the sparse solution matrix are compared with the multi-level early warning threshold. If the threshold is exceeded, the corresponding level of early warning is triggered in the three-dimensional GIS model.

[0039] For each floating-point element in the generated spatial risk weight matrix, calculate its reciprocal, and then use an extreme value normalization algorithm to linearly truncate the resulting reciprocal and map it to a preset penalty weight safety interval, such as [0.1, 1.0], to generate a spatial penalty weight matrix. Perform the matrix Hadamard product (element-by-element multiplication) to multiply the generated spatial penalty weight matrix with the regularization parameter matrix of each corresponding spatial node position in the mixed sparsity constraint term of the objective function, thus implementing non-uniform sparse constraints based on spatial prior knowledge. Develop an iterative solution loop framework based on the Alternating Direction Multiplier Method (ADMM) algorithm using native Python syntax. By employing auxiliary split variables, the originally complex unconstrained objective function is decomposed into a subproblem of primal variables, a subproblem of auxiliary variables, and a subproblem of dual multiplier updates.

[0040] In the auxiliary variable update step of the ADMM framework, a soft-threshold shrinkage operator is executed for the part corresponding to the second sparsity component. The specific steps are as follows: For each independent element in the matrix to be processed, obtain the second penalty weight threshold for the spatial location of that element; perform element-by-element shrinkage judgment, comparing the absolute value of the element with the second penalty weight threshold; if the absolute value of the element is greater than the threshold, retain the original sign of the element and subtract the second penalty weight threshold from its absolute value, using this as the updated value of the element; if the absolute value of the element is less than or equal to the threshold, directly set the updated value of the element to zero. This filters out weak, isolated noise signals.

[0041] For the overlapping group matrix block structure corresponding to the first sparse component, a block soft thresholding operator is executed. The specific steps are as follows: traverse each preset overlapping sub-matrix group, calculate the square root of the sum of the squares of all elements in the group, and use this as the overall energy value of the group; simultaneously obtain the first penalty weight threshold for the corresponding overlapping group; compare the overall energy value with the first penalty weight threshold; if the overall energy value is greater than the first penalty weight threshold, then synchronously scale all elements in the group, specifically by subtracting the ratio of the first penalty weight threshold to the overall energy value, and multiply each element in the group by this scaling ratio to obtain an updated value; if the overall energy value is less than or equal to the first penalty weight threshold, then uniformly set all elements in the group to zero. This achieves overall sparsity reduction for non-significant anomaly groups while preserving the contiguous anomaly structure in high-risk areas.

[0042] In the dual multiplier update step, the standard gradient ascent method is executed. The original residual norm and dual residual norm of two adjacent iterations are calculated. Based on the set global convergence tolerance floating-point variable, it is determined whether the KKT conditions are met. If satisfied, a `break` statement is executed to exit the `while` loop, and the converged sparse solution matrix is ​​output. Boolean indexing is used to extract outliers with absolute values ​​greater than zero in the sparse solution matrix. The `cut` function from the Pandas data analysis library is called to discretize and map these outliers to preset yellow, orange, and red three-level warning floating-point threshold boundary arrays. When the mapping result of any element is determined to be higher than the safe and normal level, a dictionary-formatted data containing the sensor ID, spatial coordinates, and warning level string is pushed to the front-end 3D GIS system in real time via Python's WebSockets asynchronous network communication library. After receiving the data stream, the front-end browser calls the Entity application interface of the Cesium.js library to parse the coordinates, locate the specific model instance, and change the rendering color of the corresponding three-dimensional entity of the alleyway to the RGBA hexadecimal highlight color value mapped to the warning level by modifying the polygon.material material property interface. Then, it calls the WebAudio API in the HTML5 standard to synthesize and play the digital audio file of the corresponding frequency of the sound and light alarm.

[0043] In some embodiments, comparing the non-zero element values ​​in the sparse solution matrix with multi-level early warning thresholds, and triggering a corresponding level of early warning in the 3D GIS model when the thresholds are exceeded, includes: Extract the non-zero elements from the sparse solution matrix, and divide the absolute values ​​of the elements into multiple anomaly levels according to a preset numerical interval mapping rule; Based on the row and column indices of the elements in the sparse solution matrix, the timestamp of the anomaly and its spatial coordinates in the 3D GIS model are obtained by reverse parsing. At the corresponding spatial coordinates of the 3D GIS model, a 3D highlighted marker with a color corresponding to the hazard level is generated, and the data panel of the associated monitoring system topologically connected to the spatial location is automatically retrieved and displayed in a pop-up window.

[0044] A global scan is performed on the sparse solution matrix output after convergence of iterative solutions to extract non-zero elements whose values ​​exceed a preset system noise floor tolerance, such as 0.05. Based on the national standard for mine disaster early warning classification, a piecewise linear numerical interval mapping dictionary is configured to discretize the absolute values ​​of the extracted elements into different levels: when the absolute value of an element is in the interval (0.05, 0.3), the element is marked as Level III, a minor abnormality warning; when it is in the interval (0.3, 0.7), it is upgraded to Level II, a moderate hazard alarm; when the absolute value exceeds the upper limit of 0.7, it is determined to be Level I, a major emergency disaster, such as detecting a strong rockburst precursor with a value of 0.85. For each non-zero element that triggers a graded alarm, the row and column indices of the element in the matrix data structure are detected.

[0045] By multiplying the row index by the sampling step size of the current warning time window (e.g., each row represents 1 second) and adding the UNIX base timestamp aligned to the start, the time node of the abnormal peak occurrence is calculated and derived, for example, 2023-10-15 14:23:45.000. The sensor spatial mapping table in memory is retrieved using the column index, and the absolute 3D coordinates of the monitoring node in the Gaussian projection of the 3D GIS environment are obtained by inversion. The underlying WebGL or Cesium rendering engine API is called to instantiate and generate halo sphere particle effects that pulsate with numerical values ​​at the corresponding coordinate point in the 3D digital twin mine as 3D highlight markers. Colors are assigned according to the anomaly level: Level I is rendered as dark red, Level II as orange-yellow, and Level III as bright yellow. The backend automatically retrieves real-time readings from associated sensors within a 50-meter radius of influence upstream and downstream of the alarm coordinate point by querying the 3D spatial topology adjacency graph. A composite data panel pops up in the sidebar of the front-end monitoring screen, displaying multi-source time-series line graphs and real-time monitoring footage from underground.

[0046] A test set was constructed using real time-series monitoring data of gas concentration and microseismic energy from a large coal mine over a continuous 30-day period. Four ablation comparison models were set up to verify the reliability of each module of the scheme: a basic early warning model using only basic absolute value sparse constraints, a first variant model with a multi-factor spatial risk weight matrix superimposed on the basic model, a second variant model with overlapping group local smoothing sparse constraints added to the first variant model, and a complete early warning model including an energy value-based penalty factor.

[0047] Experimental tests showed that the basic early warning model had an anomaly detection rate of 74.2%, a false alarm rate of 11.5%, and a spatial continuous anomaly extraction accuracy of only 65.8%. The first variant model increased the anomaly detection rate to 83.6% and reduced the false alarm rate to 7.4%. The second variant model improved the spatial continuous anomaly extraction accuracy to 89.3%. The complete early warning model showed the best overall performance, with an anomaly detection rate of 97.1%, a false alarm rate reduced to 1.2%, and a spatial continuous anomaly extraction accuracy as high as 96.5%. Figure 3 As shown, the performance differences between the base model, the two variant models, and the complete model are visualized and compared from three dimensions: detection rate, false alarm rate, and accuracy of continuous anomaly extraction.

[0048] The basic model is highly susceptible to interference from isolated environmental noise. The first variant model, by combining prior knowledge such as fault distance and goaf envelope distribution, accurately filters out irrelevant disturbances in safe areas, thus improving the early warning anti-interference capability. The second variant model, through overlapping submatrix group constraints, fully explores the spatial topological correlation between monitoring points, improving the detection capability of contiguous cluster anomaly events. The complete early warning model uses an exponential decay function to reduce the regularization weight of high-risk core areas, protecting the complete release of original anomaly characteristics in extreme energy accumulation areas. This solves the technical bottleneck of excessive smooth truncation of high-risk disaster signals due to global penalty, achieving high-fidelity and high-precision analysis of spatiotemporal disaster evolution.

[0049] One or more embodiments of the present invention also provide a multi-system collaborative early warning system for mines based on three-dimensional GIS digital twins, including the following modules: The generation module is used to acquire time-series monitoring data from multiple monitoring systems in the mine, perform standardized processing, and map it with spatial entities in the 3D GIS model to construct a spatiotemporal data matrix. Based on the pre-set fault structure, goaf, and ventilation path information in the 3D GIS model, the module calculates the prior risk weights of each spatial location and generates a spatial risk weight matrix. The calculation module is used to construct an early warning solution objective function that includes a data fidelity term and a mixed sparsity constraint term. The mixed sparsity constraint term is composed of a weighted sum of a first sparsity component and a second sparsity component. The weight coefficients between the components are initialized. After solving the problem in the current early warning cycle, the ratio of the non-zero norm contributed by each overlapping group to the non-zero norm contributed by isolated elements in the sparse solution matrix is ​​calculated. When the ratio exceeds a predetermined threshold, the weight of the second sparsity component in the next cycle is increased. For the specific overlapping group with the largest non-zero element energy in the current cycle, a penalty factor negatively correlated with the energy value is applied in the next cycle. The early warning module is used to weight the mixed sparsity constraint terms by using the reciprocals of each element in the spatial risk weight matrix, and to iteratively solve the objective function using the alternating direction multiplier method to obtain a sparse solution matrix representing the abnormal signal. The non-zero element values ​​in the sparse solution matrix are compared with multi-level early warning thresholds, and if they exceed the thresholds, the corresponding level of early warning is triggered in the three-dimensional GIS model.

[0050] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0051] The above description is merely a specific embodiment of the present invention. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A collaborative early warning method for multiple systems in mines based on 3D GIS digital twins, characterized in that, Includes the following steps: The time-series monitoring data of the mine's multi-monitoring system is acquired and standardized, and mapped with spatial entities in the 3D GIS model to construct a spatiotemporal data matrix. Based on the pre-set fault structure, goaf, and ventilation path information in the 3D GIS model, the prior risk weight of each spatial location is calculated to generate a spatial risk weight matrix. A warning solution objective function is constructed, which includes a data fidelity term and a mixed sparsity constraint term. The mixed sparsity constraint term is composed of a weighted sum of a first sparsity component and a second sparsity component. The weight coefficients between the components are initialized. After solving the current warning cycle, the ratio of the non-zero norm contributed by each overlapping group to the non-zero norm contributed by isolated elements in the sparse solution matrix is ​​calculated. When the ratio exceeds a predetermined threshold, the weight of the second sparsity component in the next cycle is increased. For the specific overlapping group with the largest non-zero element energy in the current cycle, a penalty factor negatively correlated with the energy value is applied in the next cycle. The mixed sparsity constraint term is weighted using the reciprocal of each element in the spatial risk weight matrix, and the objective function is solved iteratively using the alternating direction multiplier method to obtain a sparse solution matrix representing the abnormal signal. The non-zero element values ​​in the sparse solution matrix are compared with multi-level early warning thresholds, and if they exceed the thresholds, the corresponding level of early warning is triggered in the three-dimensional GIS model.

2. The mine multi-system collaborative early warning method based on three-dimensional GIS digital twin according to claim 1, characterized in that, The process of acquiring time-series monitoring data from the mine's multi-monitoring system, standardizing it, and mapping it to spatial entities within a 3D GIS model to construct a spatiotemporal data matrix includes: Parse the raw messages sent from each monitoring system to extract the unique sensor identifiers and measurement values; Query the preset spatial configuration database and obtain the three-dimensional spatial absolute coordinates of the corresponding sensor in the three-dimensional GIS model based on the unique identifier; An initial two-dimensional spatiotemporal grid is constructed using the total number of sensors as the column dimension of the matrix and the sampling time series within a preset time window as the row dimension of the matrix. After the measurement values ​​are standardized, they are filled into the corresponding nodes of the initial two-dimensional spatiotemporal grid according to the corresponding timestamp and the column index corresponding to the unique identifier, thereby generating the spatiotemporal data matrix.

3. The mine multi-system collaborative early warning method based on three-dimensional GIS digital twin according to claim 1, characterized in that, The method involves calculating the prior risk weights for each spatial location based on the pre-defined fault structures, goaf areas, and ventilation paths information in the 3D GIS model, generating a spatial risk weight matrix, including: Calculate the shortest geometric distance from the spatial entity to the nearest fault rupture surface, and calculate the first risk weighting factor based on the ratio of the preset hazard impact radius to the distance; Determine whether the spatial entity is located inside the closed envelope of the goaf. If it is inside, assign a preset constant greater than zero to the second risk weight factor; if it is outside, assign a value of zero. Obtain the absolute wind pressure difference between the actual monitored wind pressure and the standard wind pressure at the ventilation node where the spatial entity is located, and normalize the wind pressure difference as the third risk weighting factor. The first risk weight factor, the second risk weight factor, and the third risk weight factor are multiplied by their respective preset fixed weight coefficients and summed to obtain the prior risk weights for each spatial location, thus generating the spatial risk weight matrix.

4. The mine multi-system collaborative early warning method based on three-dimensional GIS digital twin according to claim 1, characterized in that, The construction of the early warning solution objective function includes a data fidelity term and a mixed sparsity constraint term. The mixed sparsity constraint term is composed of a weighted average of a first sparsity component and a second sparsity component, including: Subtract the preset normal security background reference matrix from the spatiotemporal data matrix to obtain the differential spatiotemporal matrix. The data fidelity term is constructed using the square of the Frobenius norm of the error between the differential spatiotemporal matrix and the sparse solution matrix to be solved. The sparse solution matrix to be solved is divided into multiple overlapping sub-matrix groups. The L2 norm of each sub-matrix group is calculated. The first sparsity component is constructed by the sum of the L2 norms of all sub-matrix groups to constrain the spatially continuous overlapping group anomalous signals. The L1 norm is calculated for each element in the sparse solution matrix to be solved, and the second sparsity component is constructed by the sum of the L1 norms of all elements to constrain isolated anomalous signals.

5. The mine multi-system collaborative early warning method based on three-dimensional GIS digital twin according to claim 1, characterized in that, The step of calculating the ratio of the non-zero norm contributed by each overlapping group to the non-zero norm contributed by isolated elements in the sparse solution matrix after solving the current early warning cycle includes: Extract the sparse solution matrix obtained in the current early warning period; Calculate the sum of squares of all non-zero elements in each overlapping group of the sparse solution matrix, and sum the calculation results of all overlapping groups to obtain the sum of the norms of the non-zero elements in the overlapping groups; Find the union of all matrix nodes covered by overlapping groups, traverse the independent non-zero elements in the sparse solution matrix that are not covered by the union, calculate the sum of squares of the independent non-zero elements, and obtain the sum of norms of isolated non-zero elements. The ratio used to adjust the weighting coefficients is obtained by dividing the sum of the norms of the non-zero elements in the overlapping group by the sum of the norms of the non-zero elements in the isolated group.

6. The mine multi-system collaborative early warning method based on three-dimensional GIS digital twin according to claim 1, characterized in that, The application of a penalty factor negatively correlated with the energy value to a specific overlapping group with the highest energy of non-zero elements in the current period in the next period includes: Calculate the sum of squares of all elements in each overlapping group in the current periodic sparse solution matrix, determine the overlapping group with the largest sum of squares as the specific overlapping group, and use the sum of squares as the energy value; A penalty factor decay function is constructed, wherein the output value of the penalty factor decay function decreases exponentially with the increase of the input energy value, and the range of the output value is limited to the interval greater than 0 and not greater than 1. The energy value of the specific overlapping group is input into the penalty factor decay function to calculate the penalty factor. When constructing the objective function for the next early warning period, the corresponding term of the specific overlapping group in the first sparsity component is multiplied by the penalty factor.

7. The mine multi-system collaborative early warning method based on three-dimensional GIS digital twin according to claim 1, characterized in that, The step of comparing the non-zero element values ​​in the sparse solution matrix with multi-level early warning thresholds, and triggering the corresponding level of early warning in the 3D GIS model when the thresholds are exceeded, includes: Extract the non-zero elements from the sparse solution matrix, and divide the absolute values ​​of the elements into multiple anomaly levels according to a preset numerical interval mapping rule; Based on the row and column indices of the elements in the sparse solution matrix, the timestamp of the anomaly and its spatial coordinates in the 3D GIS model are obtained by reverse parsing. At the corresponding spatial coordinates of the 3D GIS model, a 3D highlighted marker with a color corresponding to the hazard level is generated, and the data panel of the associated monitoring system topologically connected to the spatial location is automatically retrieved and displayed in a pop-up window.

8. A multi-system collaborative early warning system for mines based on 3D GIS digital twins, characterized in that: Includes the following modules: The generation module is used to acquire time-series monitoring data from multiple monitoring systems in the mine, perform standardized processing, and map it with spatial entities in the 3D GIS model to construct a spatiotemporal data matrix. Based on the pre-set fault structure, goaf, and ventilation path information in the 3D GIS model, the module calculates the prior risk weights of each spatial location and generates a spatial risk weight matrix. The calculation module is used to construct an early warning solution objective function that includes a data fidelity term and a mixed sparsity constraint term. The mixed sparsity constraint term is composed of a weighted sum of a first sparsity component and a second sparsity component. The weight coefficients between the components are initialized. After solving the problem in the current early warning cycle, the ratio of the non-zero norm contributed by each overlapping group to the non-zero norm contributed by isolated elements in the sparse solution matrix is ​​calculated. When the ratio exceeds a predetermined threshold, the weight of the second sparsity component in the next cycle is increased. For the specific overlapping group with the largest non-zero element energy in the current cycle, a penalty factor negatively correlated with the energy value is applied in the next cycle. The early warning module is used to weight the mixed sparsity constraint terms by using the reciprocals of each element in the spatial risk weight matrix, and to iteratively solve the objective function using the alternating direction multiplier method to obtain a sparse solution matrix representing the abnormal signal. The non-zero element values ​​in the sparse solution matrix are compared with multi-level early warning thresholds, and if they exceed the thresholds, the corresponding level of early warning is triggered in the three-dimensional GIS model.

9. The mine multi-system collaborative early warning system based on three-dimensional GIS digital twin as described in claim 8, characterized in that, The process of acquiring time-series monitoring data from the mine's multi-monitoring system, standardizing it, and mapping it to spatial entities within a 3D GIS model to construct a spatiotemporal data matrix includes: Parse the raw messages sent from each monitoring system to extract the unique sensor identifiers and measurement values; Query the preset spatial configuration database and obtain the three-dimensional spatial absolute coordinates of the corresponding sensor in the three-dimensional GIS model based on the unique identifier; An initial two-dimensional spatiotemporal grid is constructed using the total number of sensors as the column dimension of the matrix and the sampling time series within a preset time window as the row dimension of the matrix. After the measurement values ​​are standardized, they are filled into the corresponding nodes of the initial two-dimensional spatiotemporal grid according to the corresponding timestamp and the column index corresponding to the unique identifier, thereby generating the spatiotemporal data matrix.

10. The mine multi-system collaborative early warning system based on three-dimensional GIS digital twin as described in claim 8, characterized in that, The method involves calculating the prior risk weights for each spatial location based on the pre-defined fault structures, goaf areas, and ventilation paths information in the 3D GIS model, generating a spatial risk weight matrix, including: Calculate the shortest geometric distance from the spatial entity to the nearest fault rupture surface, and calculate the first risk weighting factor based on the ratio of the preset hazard impact radius to the distance; Determine whether the spatial entity is located inside the closed envelope of the goaf. If it is inside, assign a preset constant greater than zero to the second risk weight factor; if it is outside, assign a value of zero. Obtain the absolute wind pressure difference between the actual monitored wind pressure and the standard wind pressure at the ventilation node where the spatial entity is located, and normalize the wind pressure difference as the third risk weighting factor. The first risk weight factor, the second risk weight factor, and the third risk weight factor are multiplied by their respective preset fixed weight coefficients and summed to obtain the prior risk weights for each spatial location, thus generating the spatial risk weight matrix.

Images