GIS 3D Visualization System for Dynamic Evolution of Mine Gas Geology and Intelligent Evaluation of Extraction Standards

Abstract

This invention relates to the field of mine informatization and safety production management, specifically to a GIS-based 3D visualization system for dynamic evolution of mine gas geology and intelligent evaluation of drainage compliance. The system includes: a 3D dynamic modeling module for constructing a basic gas geology model and performing spatiotemporal extrapolation of gas geological parameters in unexposed areas, enabling dynamic model correction; an intelligent drainage compliance evaluation module for establishing spatial correlation between standardized drainage data and a 3D voxel grid, calculating residual gas parameters for each grid, and automatically determining the drainage compliance status of each grid according to preset rules; a 3D visualization early warning and decision-making module for dynamically displaying the results of drainage compliance status evaluation in a visual form and automatically generating drainage and borehole repair solutions for non-compliant areas; and an edge-cloud collaborative computing architecture. This invention solves the problems of lag and manual nature in drainage compliance evaluation, ultimately realizing a shift from passive response to proactive prevention in coal mine gas management.

Description

Technical Field

[0001] This invention relates to the field of mine informatization and safety production management technology, specifically to a GIS three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance. Background Technology

[0002] Gas drainage compliance evaluation is a core and crucial link in coal mine gas management and coal and gas outburst prevention, directly affecting the safety and efficiency of underground coal mine production. Currently, the field of coal mine gas drainage compliance evaluation still relies heavily on traditional technologies, resulting in several pressing industry pain points: First, geological models are static, generally using two-dimensional CAD drawings or simplified three-dimensional models to display gas geological information. This fails to accurately reflect the heterogeneous changes of gas parameters in three-dimensional space, and model updates lag behind mining progress, leading to a disconnect between decision-making and actual working conditions. Second, the drainage compliance evaluation process is discrete and manual, relying on periodic manual collection of borehole drainage data and offline calculations. This makes it difficult to achieve real-time and continuous analysis of massive amounts of data, easily creating drainage gaps and posing safety hazards. Third, there is a lack of dynamic fusion and real-time evaluation capabilities for multi-source data; existing systems lack... The existing systems lack several key features. First, they are not designed for proactive gas control. First, they are highly integrated, resulting in high deployment and maintenance costs. Second, they lack unified hardware configuration standards and data quality specifications, making them difficult for small and medium-sized mines to afford. Third, safety certification requirements for underground equipment, such as explosion-proof and environmental adaptability, are not adequately considered, leading to poor system implementation. While existing technologies include spatiotemporal graph convolutional networks and Kriging interpolation for parameter prediction, no technology has yet deeply integrated them with coal mine gas geology management and extraction compliance evaluation, nor has a standardized and implementable overall solution adapted to coal mine conditions been developed. Summary of the Invention

[0003] The present invention aims to overcome the above-mentioned defects in existing mine gas geology management and extraction compliance evaluation technologies, and to provide a GIS three-dimensional visualization intelligent evaluation system for mine gas geology dynamic evolution and extraction compliance. The core objective of this invention is to achieve integrated, automated, and visualized dynamic management of the entire process of mine gas geology description, extraction process monitoring, compliance status assessment, hazard warning, and governance optimization through the deep integration of dynamic gridded modeling, standardized management and control of multi-source data, phased spatiotemporal map algorithm iteration, edge-cloud collaborative computing, and intelligent evaluation engine. Simultaneously, it establishes unified hardware configuration standards, data quality specifications, and security certification requirements, adopts a phased implementation strategy, significantly reduces system deployment and maintenance costs, and adapts to the usage needs of mines of different sizes. Furthermore, by combining Kriging interpolation with ST-GCN and ST-GCSN algorithms in stages, it improves the prediction accuracy of gas geological parameters in unexposed areas, enabling dynamic correction of the gas geological model. It also constructs a robust residual gas parameter inversion algorithm and an intelligent borehole design mechanism to solve the problems of lag and manual intervention in extraction compliance evaluation, ultimately realizing a shift from passive response to proactive prevention in coal mine gas management.

[0004] To achieve the above objectives, the following technical solution is adopted:

[0005] This invention provides a GIS-based 3D visualization system for dynamic evolution of mine gas geology and intelligent evaluation of gas extraction compliance. The system includes: a 3D dynamic modeling module, used to discretize the target coal seam into a 3D voxel grid and construct a basic gas geology model; dynamically update the attributes of exposed grids based on mining progress data; and perform spatiotemporal extrapolation of gas geology parameters in unexposed areas to achieve dynamic model correction; and an intelligent evaluation module for gas extraction compliance, used to acquire and standardize gas extraction data; establish a spatial correlation between the standardized extraction data and the 3D voxel grid; and calculate the residual gas in each grid in real time using a built-in inversion algorithm. The system automatically determines the extraction compliance status of each grid according to preset rules based on parameters; a 3D visualization early warning and decision-making module is used to map the judgment results of the extraction compliance status to a 3D voxel grid in a visual form and display them dynamically, and automatically generate extraction and hole filling treatment plans for areas judged as non-compliant; an edge-cloud collaborative computing architecture includes mining explosion-proof edge computing nodes deployed underground and cloud computing platforms deployed on the ground. The edge computing nodes are used to perform data acquisition, preprocessing and lightweight computing tasks, and the cloud computing platform is used to perform model training and prediction calculations and 3D visualization rendering.

[0006] Furthermore, the three-dimensional dynamic modeling module includes: a basic model construction unit, used to discretize the target coal seam area into a three-dimensional voxel grid and assign initial gas geological attribute parameters to each grid; a grid dynamic activation and update unit, used to interface with the mine mining progress measurement equipment, automatically acquire mining spatial coordinate data, identify and activate the exposed grid cells based on the spatial coordinate data, and associate the on-site measured gas geological data with the corresponding activated grid cells after standardization processing to complete the attribute update of the exposed grids; and a spatiotemporal extrapolation unit, used to perform fusion calculations on the unexposed grid cells using a spatial interpolation algorithm and a spatiotemporal graph neural network algorithm, combined with the measured data of the exposed grids, to obtain predicted values ​​of gas geological parameters in the unexposed area, and dynamically correct the three-dimensional grid model based on the predicted values.

[0007] Furthermore, the spatiotemporal extrapolation unit adopts a phased iterative strategy. In the first phase, a spatiotemporal graph convolutional network is used for prediction. In the second phase, after verification with field data, it is upgraded to a spatiotemporal graph complementary scattering network for prediction. The spatiotemporal extrapolation unit performs a weighted fusion of the calculation results of the spatial interpolation algorithm and the prediction results of the spatiotemporal graph neural network algorithm. The fusion weight is dynamically adjusted according to the historical prediction error to obtain the final predicted value of the gas geological parameters in the unexposed area. The spatial interpolation algorithm adopts the Kriging interpolation algorithm, and its calculation task is executed at the downhole edge computing node. The prediction calculation task of the spatiotemporal graph neural network algorithm is executed on the cloud computing platform.

[0008] Furthermore, the spatiotemporal graph complementary scattering network uses graph complementary scattering layers as its basic modules. Each graph complementary scattering layer contains a fixed feature extraction branch and a trainable complementary feature extraction branch. The fixed feature extraction branch is generated by a spatiotemporal graph scattering transform optimized by pruning techniques, and extracts the spatiotemporal features of gas geological parameters through mathematically designed spatial and temporal wavelet filters. The trainable complementary feature extraction branch adopts a filter structure complementary to the fixed feature extraction branch, including trainable spatial and temporal complementary filters, and generates a trainable graph shift matrix through a surrogate parameter training mechanism. The surrogate parameter training mechanism obtains a trainable matrix that conforms to the characteristics of a Markov matrix after row normalization of the surrogate parameter matrix. The spatiotemporal graph complementary scattering network is trained using a hybrid optimization strategy, employing an adaptive moment estimation optimizer in the early stage and a stochastic gradient descent optimizer for fine-tuning in the later stage. The loss function combines mean squared error and L1 regularization. The spatiotemporal graph complementary scattering network adopts an online update mechanism that combines incremental learning and periodic retraining. The incremental learning cycle is a preset number of days, during which small-batch fine-tuning is performed on the original model. The periodic retraining cycle is a preset number of months, during which the model is fully retrained by incorporating all newly added data.

[0009] Furthermore, the intelligent evaluation module for achieving extraction standards includes: an extraction data access and processing unit, used to interface with the downhole gas extraction monitoring system, automatically acquire extraction flow rate, gas concentration, and extraction duration data of the extraction borehole, and perform outlier filtering and missing value completion processing on the acquired data according to a preset data quality threshold; a grid-bore intelligent association unit, used to establish the association relationship between each grid cell and the extraction boreholes within its influence range based on the spatial topological relationship between the borehole spatial coordinates and the three-dimensional voxel grid, and determine the extraction influence weight of each borehole on the grid according to the spatial distance between the borehole and the grid; and a residual gas parameter inversion calculation unit, used to calculate the residual gas parameter inversion based on the original gas content of each grid cell, the original... The system uses the initial gas pressure, along with the cumulative extraction flow rate, average gas concentration, and extraction duration of the associated extraction boreholes, combined with the grid coal seam volume, apparent coal density, and gas extraction utilization coefficient, to calculate the residual gas content and residual gas pressure of each grid unit in real time using a built-in inversion algorithm. The inversion algorithm incorporates a data fault tolerance mechanism; when the extraction data of a single grid is incomplete, effective extraction data from surrounding grids is used for collaborative calculation. An automatic compliance determination unit automatically compares the residual gas content and residual gas pressure obtained from the inversion calculation with the extraction compliance threshold specified in industry standards. Following a preset four-level fault-tolerant determination rule, each grid unit is classified into one of four states: compliant, critical warning, non-compliant, or control blind zone.

[0010] Furthermore, the four-level fault-tolerant judgment rule adopted by the automatic judgment unit for the compliance status is as follows: when the residual gas parameter is lower than 80% of the critical value for drainage compliance and the associated drainage data is complete and valid, the grid unit is judged to be in compliance status; when the residual gas parameter is between 80% and 100% of the critical value for drainage compliance and the associated drainage data is complete and valid, the grid unit is judged to be in critical warning status; when the residual gas parameter is higher than the critical value for drainage compliance and the associated drainage data is complete and valid, the grid unit is judged to be in non-compliance status; when the grid unit has no effective drainage borehole coverage, or the associated drainage data and measured data are abnormally missing and cannot be supplemented, or the inversion calculation result exceeds the preset reasonable range, the grid unit is judged to be in control blind zone status.

[0011] Furthermore, the three-dimensional visualization early warning and decision-making module includes: a four-color early warning visualization unit, used to map the compliance status judgment results of each grid output by the intelligent evaluation module for compliance of extraction to a three-dimensional voxel grid in real time, and to use green, yellow, red and gray to correspond to compliance, critical warning, non-compliance and control blind zone status in the GIS three-dimensional scene, forming a dynamically updated four-color early warning color cloud map; and an intelligent hole filling design unit, used to automatically generate an optimized design scheme for extraction hole filling for continuous grid areas judged as non-compliance and control blind zone, combined with mine construction specifications and geological conditions.

[0012] Furthermore, the four-color early warning visualization unit adopts an edge-cloud collaborative update mechanism, in which data changes are pushed in real time by the downhole edge node and lightweight rendering is performed in the cloud to achieve second-level dynamic refresh of the chromatogram cloud map.

[0013] Furthermore, the intelligent borehole repair design unit adopts a step-by-step refined design process, including: dividing the continuous non-compliant and control blind zone grid into several borehole repair sub-regions according to geological structure, mining face division, and borehole accessibility; preliminarily screening the basic parameters for borehole repair based on the coal seam conditions and residual gas parameters of each sub-region; optimizing the borehole location based on a spatial uniform coverage algorithm to ensure that the extraction influence range of adjacent boreholes meets the preset overlap rate requirements, and that the distance between the borehole location and existing roadways and existing boreholes complies with construction safety specifications; outputting a borehole repair scheme containing borehole location coordinates, borehole parameters, and construction suggestions, and simulating and displaying the extraction coverage range after borehole repair in a GIS 3D scene.

[0014] Furthermore, the edge-cloud collaborative update mechanism adopted by the four-color early warning visualization unit includes: the underground edge node only pushes the extraction compliance status data of the grid cells that have changed; the cloud platform locally refreshes the rendering attributes of the corresponding grid cells in the three-dimensional scene based on the incremental update algorithm, without reloading and rendering the entire three-dimensional scene; the cloud platform has a built-in GIS offline rendering engine, which supports the continuous provision of three-dimensional visualization viewing function based on the last synchronized full status data in the event of a mine network interruption, and automatically synchronizes data and completes status with the edge node after the network is restored.

[0015] Compared with the prior art, the present invention achieves the following beneficial effects:

[0016] 1. This invention discretizes the target coal seam into a three-dimensional voxel grid and constructs a basic model. Combined with the mining advance, it realizes automatic and real-time updates of grid attributes. It integrates Kriging interpolation and a phased intelligent algorithm to complete the spatiotemporal extrapolation of gas geological parameters, which completely changes the shortcomings of the traditional static two-dimensional model. It accurately reflects the heterogeneous changes of gas geological parameters in three-dimensional space, and the model fidelity and timeliness are greatly improved, providing accurate and real-time basis for gas control decisions.

[0017] 2. This invention constructs a robust residual gas parameter inversion algorithm, realizing intelligent correlation between extraction data and three-dimensional grid, and real-time calculation of residual gas content and pressure. At the same time, it formulates a four-level fault-tolerant compliance judgment rule to avoid the subjective error and lag of manual evaluation, realizes the automated and hierarchical judgment of extraction compliance status, effectively solves the industry problem of difficulty in finding extraction blank zones, and improves the accuracy and reliability of gas extraction compliance evaluation.

[0018] 3. This invention dynamically displays the results of gas extraction compliance assessment in a four-color early warning chromatogram cloud map within a GIS 3D scene, achieving panoramic and transparent monitoring of underground gas extraction status, making potential hazard areas readily apparent. Simultaneously, for non-compliant areas and control blind spots, it generates directly implementable optimized design schemes for gas extraction and hole filling based on coal mine on-site construction specifications. Furthermore, it achieves high-precision prediction of gas parameters in unexposed areas through a phased spatiotemporal map algorithm, endowing the system with forward-looking prediction and intelligent decision support capabilities, realizing the transformation of gas management from passive response to proactive prevention and control.

[0019] 4. This invention adopts a three-tiered, phased implementation strategy, designing three versions: Basic, Advanced, and High-end, to meet the needs of small and medium-sized mines, mines with information technology infrastructure, and large-scale modern mines, respectively. Mines can gradually upgrade their functions according to their own development, avoiding high one-time costs. At the same time, it constructs a collaborative computing architecture of "underground mine explosion-proof edge nodes + ground cloud platform," which pushes lightweight computing to the edge nodes, reduces the pressure on cloud computing and transmission, further reduces system operation and maintenance costs, and allows mines of different sizes to enjoy the benefits of intelligent gas management technology.

[0020] 5. This invention clarifies the hardware configuration standards for underground edge nodes. All underground equipment complies with coal mine safety regulations and obtains mining product safety marks (MA) and explosion-proof certificates. The network architecture adopts a physical isolation design to meet the environmental adaptability requirements of underground explosion-proof, anti-interference, dustproof, and waterproof. At the same time, it formulates unified specifications for gas geological data acquisition, labeling, and quality control, and provides a supporting on-site operator training system to ensure that the various technical solutions of the system can be deeply adapted to the working conditions of the coal mine site, greatly improving the practicality and operability of the system.

[0021] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0022] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the invention. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0023] Figure 1 This is a schematic diagram of the architecture of the GIS three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance according to an embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of the spatiotemporal extrapolation of gas geological parameters in an embodiment of the present invention, which integrates Kriging interpolation and a phased spatiotemporal mapping algorithm.

[0025] Figure 3 This is a schematic diagram of residual gas parameter inversion and intelligent evaluation of extraction compliance in an embodiment of the present invention;

[0026] Figure 4 This is a visual rendering illustration of an embodiment of the present invention;

[0027] Figure 5 This is a simulation result comparing the loss values ​​of ST-GCSN and ST-GCN under training / test data according to the present invention. Detailed Implementation

[0028] 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.

[0029] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0030] This invention aims to overcome the problems of existing mine gas geology management and extraction compliance evaluation technologies, such as static models, manual evaluation, lagging data fusion, weak decision support, unclear algorithm parameters, lack of standard edge node configuration, lack of specific thresholds for data control, and insufficient system security adaptability. It provides a Geographic Information System (GIS) 3D visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance, which is phased and implementable, has strong algorithm robustness, high deployment cost adaptability, and standardized and reproducible parameters. This invention achieves integrated, automated, and visualized dynamic management of the entire process, including mine gas geology description, extraction process monitoring, compliance status evaluation, hazard warning, and governance optimization, through dynamic grid modeling, standardized multi-source data control, phased spatiotemporal map algorithm iteration, edge-cloud collaborative computing, and intelligent evaluation engine construction. It also considers the deployment needs of mines of different sizes, clearly defines all core parameters, hardware configurations, and safety specifications, avoids practical risks in technology implementation, and complies with the full-process standard requirements for coal mine safety production. The core technical solution of this invention is as follows:

[0031] Figure 1This is a schematic diagram of the architecture of the GIS three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance according to an embodiment of the present invention. Figure 1 As shown, a GIS-based 3D visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance includes:

[0032] The 3D dynamic modeling module 110 is used to discretize the target coal seam into a 3D voxel grid and construct a basic gas geological model. It dynamically updates the properties of the exposed grid based on mining progress data and performs spatiotemporal extrapolation of the gas geological parameters of the unexposed area to achieve dynamic correction of the model.

[0033] Furthermore, the 3D dynamic modeling module 110 includes: a basic model construction unit, used to discretize the target coal seam area into a 3D voxel grid and assign initial gas geological attribute parameters to each grid; a grid dynamic activation and update unit, used to interface with the mine mining progress measurement equipment, automatically acquire mining spatial coordinate data, identify and activate the exposed grid cells based on the spatial coordinate data, and associate the on-site measured gas geological data with the corresponding activated grid cells after standardization processing to complete the attribute update of the exposed grids; and a spatiotemporal extrapolation unit, used to perform fusion calculations on the unexposed grid cells using a spatial interpolation algorithm and a spatiotemporal graph neural network algorithm, combined with the measured data of the exposed grids, to obtain predicted values ​​of gas geological parameters in the unexposed area, and dynamically correct the 3D grid model based on the predicted values. More specifically, the 3D dynamic modeling module 110 is implemented through the following steps:

[0034] S1: Construct 3D dynamic modeling module 110

[0035] Step S1 uses a three-dimensional voxel mesh as the basic unit, combined with dynamic updates triggered by mining advance, and integrates Kriging interpolation and a phased spatiotemporal graph algorithm to achieve high-precision spatiotemporal prediction and dynamic extrapolation of gas geological parameters. An edge-cloud collaborative computing architecture is introduced to reduce cloud pressure, and specific thresholds for data standardization are set. At the same time, a phased iterative strategy is adopted, first verifying with a spatio-temporal graph convolutional network (ST-GCN) and then upgrading with a spatio-temporal graph complementary scattering network (ST-GCSN). The training parameters and model update mechanism of ST-GCSN are clarified, providing high-precision support for the dynamic correction of the gas geological model.

[0036] S11: Construction of Basic Gas Geological Model

[0037] The target coal seam region is discretized into a three-dimensional voxel grid (X×Y×Z) with flexibly configurable resolution. The recommended resolution is 0.5m×0.5m×0.5m to 5m×5m×5m, with each grid serving as an independent attribute-bearing unit. Basic gas geological attribute parameters such as geological structure, coal seam thickness, coal seam hardness, original gas content, and original gas pressure are initially assigned to each grid to construct a three-dimensional gridded basic model of gas geology (hereinafter referred to as the gas geological basic model), which serves as a standardized carrier for subsequent dynamic updates and deductions.

[0038] S12: Mesh Dynamic Activation and Normalized Attribute Update Driven by Mining Footage

[0039] S121: Standardized Integration and Grid Activation of Mining Footage Data

[0040] The system is integrated with standardized interfaces for mining operation planning systems, laser scanning / measuring robots, and other advance measurement equipment. Communication protocol adaptation is achieved through coordination with equipment manufacturers, enabling daily automatic acquisition of spatial coordinates and actual advance data of the mining face. Based on the acquired spatial coordinates, the system automatically identifies and activates the exposed grid cells in the three-dimensional coal seam grid.

[0041] S122: Standardized Management and Attribute Updating of Gas Geological Data

[0042] (1) Formulate specifications for gas geological data collection, labeling and quality control, clarify specific outlier thresholds, missing value thresholds and standardized processing procedures, and provide supporting operational training systems for mine field data collection personnel. The core data quality thresholds are: ① Gas concentration: <0.1% or >100% is considered an outlier; ② Drainage flow rate: <0m³ / min or exceeding the rated maximum flow rate of the mine drainage equipment is considered an outlier; ③ Gas content / pressure: deviation from the mean of the exposed grid in the area ±3σ is considered an outlier; ④ Data missing: no valid data uploaded for 10 consecutive minutes at a single monitoring point is considered a missing value. (2) Standardize the field measured data (geological logging data, direct gas content measurement, gas pressure measurement, etc.) according to the above specifications. Outliers are replaced by interpolation of valid values ​​in the same time period of adjacent grids. Missing values ​​are filled by time series interpolation (dynamic drainage data) or spatial interpolation (static geological data) according to the data type. (3) The standardized measured data are automatically associated with the corresponding activated grid cells to complete the real-time update of the gas geological properties of the exposed grid, ensuring the quality and consistency of the model input data.

[0043] S13: Spatiotemporal extrapolation of gas geological parameters by integrating Kriging interpolation and staged spatiotemporal mapping algorithms

[0044] like Figure 2The diagram shown is a schematic representation of the spatiotemporal extrapolation of gas geological parameters using a kriging interpolation and staged spatiotemporal mapping algorithm, as described in an embodiment of the present invention. The specific steps include the following:

[0045] S131: Kriging space interpolation optimization calculation

[0046] For unexposed grid cells, ordinary kriging interpolation is used for preliminary interpolation estimation. An anomaly detection mechanism is introduced during the interpolation process to remove outlier samples that deviate from the threshold, improving the robustness of the interpolation results and enabling the first dynamic correction of the gas geological foundation model. This calculation task is completed down to the downhole edge node. The core calculation formulas and symbols are defined as follows:

[0047] Mutation function: ;

[0048] Kriging interpolation estimate: ;

[0049] in, : The variogram is used to characterize the degree of variation of gas geological property values ​​with spatial distance of sample points; Spatial distance between sample points, i.e., the spatial distance between two gas geological sample points involved in the interpolation calculation; Distance is The number of sample point pairs, i.e., all spatial distances equal to The total number of gas geological sample point pairs; : The gas geological property values ​​of the exposed grid, that is, the measured values ​​of properties such as gas content and gas pressure corresponding to the three-dimensional voxel grid that has been mined and exposed underground; : Interpolated estimates of unexposed grids, i.e., gas geological property estimates of unexcavated grids in the mine obtained by Kriging interpolation; Interpolation weight, which is the weight coefficient of each exposed grid sample point in the Kriging interpolation calculation; : The number of neighboring sample points involved in the interpolation, i.e., the number of surrounding exposed grid sample points selected for calculating the attribute values ​​of unexposed grids. A value of 8 to 12 is recommended. The weight constraint for Kriging interpolation is that the sum of the weights of all sample points involved in the interpolation is 1, which ensures the reasonableness of the interpolation result.

[0050] S132: Prediction Optimization of the Staged Spatiotemporal Graph Algorithm

[0051] The spatiotemporal extrapolation unit adopts a phased iterative strategy. In the first phase, a spatiotemporal graph convolutional network is used for prediction. In the second phase, after verification with field data, it is upgraded to a spatiotemporal graph complementary scattering network for prediction. The spatiotemporal extrapolation unit performs a weighted fusion of the calculation results of the spatial interpolation algorithm and the prediction results of the spatiotemporal graph neural network algorithm. The fusion weight is dynamically adjusted according to the historical prediction error to obtain the final predicted values ​​of gas geological parameters in the unexposed area. The calculation task of the spatial interpolation algorithm is executed at the downhole edge computing node, while the prediction calculation task of the spatiotemporal graph neural network algorithm is executed on the cloud computing platform.

[0052] A phased iterative strategy of first ST-GCN and then ST-GCSN is adopted. The effectiveness of the prediction process is first verified through the mature ST-GCN. After the field application is stable, the effect is verified by small sample data in the mine before upgrading to ST-GCSN. At the same time, an edge-cloud collaborative computing architecture is introduced, which pushes lightweight computing such as data preprocessing to underground edge nodes, and only uploads feature data to the cloud for deep prediction calculation, reducing the computing and transmission pressure on the cloud.

[0053] (1) First stage: ST-GCN model prediction (basic validation version)

[0054] ① Model structure: A two-layer network structure is adopted, which integrates spatial graph convolution and temporal convolution. The spatial layer constructs graph convolution kernels based on the adjacency matrix to capture the spatial correlation of the grid. The temporal layer uses 1D convolution to capture the temporal evolution characteristics of gas parameters. Finally, the prediction results are output through a fully connected layer.

[0055] ② Model training: The input consists of a historical mining progress sequence of the mine and a spatiotemporal map signal constructed from standardized gas geological parameter spatiotemporal distribution data. ,in This represents the number of nodes in the spatial grid. The time step is denoted as ; the output is the spatiotemporal prediction of gas content and gas pressure for each grid cell in the unexposed area; the training parameters are set to a learning rate of 0.001, a batch size of 32, and 100 training epochs, using the Adam (Adaptive Moment Estimation) optimizer and the mean squared error (MSE) loss function; the training data sources are the mine's historical mining ledger, standardized gas geological exploration data, and historical data from the underground real-time monitoring system.

[0056] (2) Second stage: ST-GCSN model prediction (accuracy upgrade version)

[0057] The spatiotemporal graph complementary scattering network uses graph complementary scattering layers as its basic modules. Each graph complementary scattering layer contains a fixed feature extraction branch and a trainable complementary feature extraction branch. The fixed feature extraction branch is generated by a spatiotemporal graph scattering transform optimized by pruning techniques, and extracts the spatiotemporal features of gas geological parameters through mathematically designed spatial and temporal wavelet filters. The trainable complementary feature extraction branch adopts a filter structure complementary to the fixed feature extraction branch, including trainable spatial and temporal complementary filters. A trainable graph shift matrix is ​​generated through a surrogate parameter training mechanism. The training mechanism involves row-oriented normalization of the surrogate parameter matrix to obtain a trainable matrix that conforms to the characteristics of a Markov matrix. The spatiotemporal graph complementary scattering network is trained using a hybrid optimization strategy, employing an adaptive moment estimation optimizer in the early stages and a stochastic gradient descent optimizer for fine-tuning in the later stages. The loss function combines mean squared error and L1 regularization. The spatiotemporal graph complementary scattering network uses an online update mechanism combining incremental learning and periodic retraining. The incremental learning cycle is a preset number of days, performing small-batch fine-tuning based on the original model. The periodic retraining cycle is a preset number of months, incorporating newly added full data for complete retraining. The specific implementation process is as follows:

[0058] Model Structure: This invention designs a Graph Complementary Scattering Layer (GCSL) as the basic module of ST-GCSN. Each GCSL contains two complementary components: a fixed tree node and a trainable tree node. The fixed tree node is generated by a Spatio-Temporal Graph Scattering Transform (ST-GST) optimized by pruning techniques, i.e., a spatio-temporal graph complementary scattering network. ST-GST is a mathematically interpretable feature extraction method for spatio-temporal graph signals. Its core principle is to generate a tree-structured multi-scale feature representation by iteratively applying a mathematically designed spatio-temporal graph wavelet filter and a nonlinear activation function. The pruning threshold of ST-GST is set to 0.002 to retain effective feature nodes, and the mathematically designed spatio-temporal graph wavelet filter is used to generate the feature representation. , It captures key spatiotemporal information of gas geological parameters; trainable tree nodes are constructed in a complementary form to fixed tree nodes, employing... and Complementary filter design, , As a trainable space / time graph shift matrix, its Markov matrix property is guaranteed through a surrogate parameter training mechanism, i.e., the surrogate parameter matrix... Horizontal softmax normalization Obtain a trainable matrix and train only. To avoid the matrix property failure problem of direct training; a 2-layer GCSL is set up, with a spatial scale number of Time scale number After multiple iterations and stacking, all tree node features are concatenated into the model output features. The predictor uses a single hidden layer multi-layer perceptron (MLP) with 128 hidden layer neurons.

[0059] in, Spatial dimension spatiotemporal map wavelet filter, used to capture the main spatiotemporal information of gas geological parameters; : Time-dimensional spatiotemporal wavelet filter, used to capture the main spatiotemporal information of gas geological parameters in the time dimension; Spatial scale, the spatial dimension scale parameter of the spatiotemporal wavelet filter; Time scale: The time dimension scale parameter of the spatiotemporal graph wavelet filter; Spatial map shift matrix, used to characterize the spatial relationships between nodes in the gas geology grid; : Time-shift matrix, used to characterize the temporal evolution correlation of gas geological parameters; Trainable spatial complementary filters complement the spatial filters of fixed tree nodes to capture secondary spatial information of gas geological parameters; The identity matrix ensures the integrity and rationality of the filtering operation. Trainable temporal complementary filters complement the temporal filters of fixed tree nodes to capture secondary temporal information of gas geological parameters; : The transpose of the filter is adapted to the complementary computation logic of trainable tree nodes. : Trainable spatial graph shift matrix, which guarantees the Markov matrix properties through a surrogate parameter training mechanism; : Trainable time-map shift matrix, which ensures Markov matrix properties through a surrogate parameter training mechanism; : Proxy parameter matrix; Trainable graph shift matrix; This is a row-oriented softmax normalization function used to normalize the data. Transform into a trainable matrix that conforms to the properties of a Markov matrix. ; : Number of spatial scales, the total number of spatial dimension scales set in the ST-GCSN model; : Number of time scales, the total number of time dimension scales set in the ST-GCSN model.

[0060] ② Model Training: The input is consistent with ST-GCN, and the output is a high-precision prediction of gas parameters in unexposed areas; the optimizer adopts a hybrid optimization strategy of Adaptive Moment Estimation (Adam) + Stochastic Gradient Descent (SGD), with Adam optimization in the early stage (learning rate... Weight decay Later, SGD fine-tuning was performed (learning rate 0.0001, momentum 0.9); the loss function used was mean squared error (MSE) + L1 regularization, with the following formula:

[0061] ;

[0062] The training batch size is 32, the number of training rounds is 200, and an early stopping method (patience=20) is used to prevent overfitting. The training data consists of standardized multi-source historical data from the mine. The model is fine-tuned based on the features trained by ST-GCN to reduce training costs. Before the model is deployed, it needs to be validated with a small sample of field-measured data from the mine, and the prediction error is below [value missing]. Only then can it be put into on-site use.

[0063] The first moment estimate of the Adam optimizer is the exponential decay rate, used to calculate the exponential moving average of the gradient; : The second moment estimate of the Adam optimizer is the exponential decay rate, used to calculate the exponential moving average of the squared gradient; Loss: The loss function of the ST-GCSN model, consisting of mean squared error and L1 regularization; L1 regularization coefficient, used to balance the mean squared error loss and the weight regularization term, is preferred. ; Model weights: The weight matrix / vector of all trainable parameters in the ST-GCSN model; Model predictions: Predicted gas geological parameters for unexposed areas from ST-GCSN output. : Measured values, actual values ​​of gas geological parameters collected on-site in the mine.

[0064] ③ Model update mechanism: An online update method combining incremental learning and periodic retraining is adopted. Incremental learning involves collecting new standardized measured data from the mine site every 7 days and making small-batch fine-tuning on the original model (batch size 16, learning rate 0.00001) to adapt to the dynamic changes in gas geological parameters. Periodic retraining involves integrating all newly added data every 3 months to perform a complete retraining of the model, update the model weights, and ensure long-term prediction accuracy. All update processes are completed in the cloud, and the updated model is lightweighted and distributed to edge nodes.

[0065] (3) Model fusion application: The Kriging interpolation results are weighted and fused with the prediction results of the spatiotemporal graph algorithm (ST-GCN / ST-GCSN). The weights are dynamically adjusted according to the historical prediction errors. The smaller the prediction error, the higher the weight. The accurate prediction values ​​of gas content and pressure of the unexposed area grid are obtained, realizing the forward-looking dynamic extrapolation of the three-dimensional gas geological model and completing the second correction of the model.

[0066] The intelligent evaluation module 120 for gas extraction compliance is used to acquire and standardize gas extraction data, establish a spatial relationship between the standardized extraction data and the three-dimensional voxel grid, calculate the residual gas parameters of each grid in real time through the built-in inversion algorithm, and automatically determine the extraction compliance status of each grid according to preset rules.

[0067] The intelligent evaluation module 120 for achieving residual gas parameter inversion and intelligent evaluation of drainage compliance is used. This module includes: a drainage data access and processing unit, used to interface with the downhole gas drainage monitoring system, automatically acquiring drainage flow rate, gas concentration, and drainage duration data from the drainage boreholes, and performing outlier filtering and missing value completion processing on the acquired data according to preset data quality thresholds; a grid-borehole intelligent association unit, used to establish the association relationship between each grid cell and the drainage boreholes within its influence range based on the spatial topological relationship between the borehole spatial coordinates and the three-dimensional voxel grid, and to determine the drainage influence weight of each borehole on the grid based on the spatial distance between the borehole and the grid; and a residual gas parameter inversion calculation unit, used to calculate the original gas content and original gas pressure of each grid cell. The system calculates the residual gas content and residual gas pressure of each grid unit in real time using a built-in inversion algorithm, based on the cumulative drainage flow rate, average gas concentration, and drainage duration of the associated drainage boreholes, combined with the grid coal seam volume, apparent coal density, and gas drainage utilization coefficient. The inversion algorithm incorporates a data fault tolerance mechanism; when the drainage data of a single grid is incomplete, effective drainage data from surrounding grids is used for collaborative calculation. An automatic compliance status determination unit automatically compares the residual gas content and residual gas pressure obtained from the inversion calculation with the drainage compliance threshold values ​​specified in national or industry standards. According to a preset four-level fault-tolerant determination rule, each grid unit is determined to be in one of four states: compliant, critical warning, non-compliant, or control blind zone. The intelligent drainage compliance evaluation module 120 is implemented through the following steps:

[0068] S2: Construct an intelligent evaluation module for sampling compliance 120

[0069] Step S2 constructs a robust inversion calculation engine with built-in core algorithms and data fault tolerance mechanisms that conform to national standards. It strictly follows the quantity and quality thresholds in step S122 to perform preprocessing of the extracted data, thereby realizing the association between the extracted data and the three-dimensional grid, real-time inversion of residual gas parameters, and automatic determination of whether the extraction meets the standards. Figure 3 This is a schematic diagram illustrating the residual gas parameter inversion and intelligent evaluation of extraction compliance in an embodiment of the present invention. Step S2 specifically includes the following steps:

[0070] S21: Standardized access and intelligent grid correlation of extracted data

[0071] S211: Standardization and preprocessing of sampled data

[0072] The system is standardized and connected to the downhole gas extraction monitoring system to automatically acquire real-time / cumulative extraction data of all gas extraction boreholes, including extraction flow rate, gas concentration, and extraction duration. Outlier filtering and missing value completion are performed according to the core data quality threshold in step S122. Short-term missing data are completed using time series interpolation. Data missing for more than 30 minutes is marked as data anomaly.

[0073] S212: Intelligent correlation between extracted data and 3D mesh

[0074] Based on the spatial topological relationship between borehole spatial coordinates and coal seam 3D grids, a spatial neighborhood search algorithm is used to automatically establish the association between each 3D grid and all drainage boreholes affecting the area, clarifying the effective drainage borehole coverage and drainage influence weight of each grid. The drainage influence weight is determined by weighting the spatial distance between the borehole and the grid, providing accurate drainage data support for inversion calculation.

[0075] S22: Robust real-time inversion calculation of residual gas content and pressure

[0076] The intelligent logic operation engine for achieving coal mine gas drainage standards incorporates the core calculation algorithm from the "Interim Provisions on Coal Mine Gas Drainage Standards" and introduces a data fault tolerance mechanism. When the drainage data of a single grid is incomplete, it uses the valid drainage data from surrounding grids for collaborative calculation. The core inversion calculation formula is:

[0077]

[0078] in, Residual gas content in the grid (unit: (This refers to the remaining gas content within the three-dimensional voxel grid of the coal seam after gas extraction.) Original gas content of the grid (unit: (This refers to the original gas content of the coal seam before gas extraction was carried out using a three-dimensional voxel grid.) The average pure flow rate of the associated borehole (unit: The average flow rate (T) refers to the average flow rate of all drainage boreholes associated with the coal seam grid during the drainage duration T, calculated as follows: , Cumulative drainage flow rate of associated boreholes (unit: (), refers to the cumulative gas extraction volume of all gas extraction boreholes associated with the coal seam grid; The average gas concentration in associated boreholes (unit: %, calculated after removing outliers) refers to the average gas concentration of all extraction boreholes associated with this coal seam grid after removing outliers. Sampling duration (unit: ), refers to the total effective gas drainage time of the drainage boreholes associated with the coal seam grid; The volume of the grid coal seam (unit: (), refers to the actual coal seam volume of the three-dimensional voxel grid; Apparent density of coal (unit: (Field measurement), refers to the apparent density value of coal within the grid of the coal seam obtained through field measurement; This is the gas extraction and utilization coefficient, which has no unit and is determined based on actual mine measurement data. ; Residual gas pressure in the grid (unit: MPa) refers to the residual gas pressure within the three-dimensional voxel grid of the coal seam after gas extraction. The gas content-pressure fitting coefficient, in units of Based on laboratory tests of coal seams, the goodness of fit was determined. .

[0079] Based on the above formula, and combined with the coal reserves and original gas parameters corresponding to each three-dimensional grid, the engine performs real-time inversion calculation of the residual gas content and residual gas pressure of each grid unit.

[0080] S23: Automatic Classification of Sampling Compliance Status

[0081] The engine automatically compares the residual gas content and residual gas pressure obtained from the inversion with the extraction compliance thresholds specified in national standards. It then determines the extraction compliance status of each grid cell according to a preset four-level fault-tolerant judgment rule. This rule fully considers data anomalies to avoid misjudgments. Specifically:

[0082] (1) Meets the standard: The residual gas parameters are below 80% of the critical value, and the extraction data are complete and valid;

[0083] (2) Critical warning: The residual gas parameters are between 80% and 100% of the critical value, and the extraction data are complete and valid;

[0084] (3) Not up to standard (exceeding standard): The residual gas parameters are higher than the critical value, and the sampling data is complete and valid;

[0085] (4) Control blind zone: There is no effective extraction borehole coverage, or the extraction / measurement data is abnormal / missing and cannot be supplemented, or the parameter judgment results are outside the reasonable range.

[0086] The 3D visualization early warning and decision-making module 130 is used to map the judgment result of the extraction compliance status to a 3D voxel grid in a visual form and display it dynamically. It automatically generates extraction and hole filling treatment plan for areas judged as non-compliant.

[0087] Furthermore, the 3D visualization early warning and decision-making module 130 is used to realize 3D visualization early warning and intelligent decision-making for extraction management in a Geographic Information System (GIS). It includes: a four-color early warning visualization unit, used to map the compliance status judgment results of each grid output by the intelligent evaluation module for extraction compliance to a 3D voxel grid in real time. In the GIS 3D scene, green, yellow, red, and gray correspond to compliance, critical warning, non-compliance, and control blind zone states, respectively, forming a dynamically updated four-color early warning chromatogram; and an intelligent borehole design unit, used to automatically generate optimized borehole design schemes for continuous grid areas judged as non-compliance and control blind zones, combined with mine construction specifications and geological conditions. The four-color early warning visualization unit adopts an edge-cloud collaborative update mechanism, with real-time data changes pushed by underground edge nodes and lightweight rendering performed in the cloud, achieving second-level dynamic refresh of the chromatogram. Specifically, this is achieved through the following steps:

[0088] Step S3: Construct a 3D visualization early warning and decision-making module 130

[0089] Step S3 refines the implementation process and constraints of the hole filling design, formulates an executable hole filling plan in conjunction with the on-site construction specifications of the mine, and relies on the edge-cloud collaborative architecture to realize the dynamic visualization of the extraction results, ensuring the real-time updating of the visualization.

[0090] S31: Dynamic visualization of four-color early warning chromatogram cloud map

[0091] The four-color early warning visualization unit of the 3D visualization early warning and decision-making module 130 adopts an edge-cloud collaborative update mechanism, which includes: the underground edge nodes only push the extraction compliance status data of the grid units that have changed; the cloud platform locally refreshes the rendering attributes of the corresponding grid units in the 3D scene based on an incremental update algorithm, without reloading and rendering the entire 3D scene; the cloud platform has a built-in GIS offline rendering engine, which supports continuous 3D visualization viewing based on the last synchronized full status data in the event of a mine network interruption, and automatically synchronizes data and completes status with the edge nodes after the network is restored. The specific process is as follows:

[0092] The extraction compliance status judgment result of step S23 is mapped in real time to the corresponding grid cells of the three-dimensional meshed model of the coal seam. In the GIS three-dimensional scene, it is visualized and rendered in the form of a four-color early warning chromatogram cloud map. The chromatogram mapping rules correspond one-to-one with the judgment rules: green represents the compliant area, yellow represents the critical warning area, red represents the non-compliant (exceeding standard) area, and gray represents the extraction control blind zone. An edge-cloud collaborative update mechanism is adopted, with underground edge nodes pushing data changes in real time, and the cloud performing only lightweight rendering, achieving second-level dynamic updates of the chromatogram cloud map as mining progresses, extraction data is updated, and gas geological parameters are extrapolated. The GIS platform conforms to coal mine data visualization specifications, supports offline caching to avoid visualization failure due to network interruptions, and supports multi-view, multi-scale viewing, adapting to the usage needs of the underground site and the ground dispatch center. Figure 4 The image shown is a visual rendering illustration of an embodiment of the present invention.

[0093] S32: Step-by-step intelligent optimization design for drilling and filling boreholes in substandard areas

[0094] For the red non-compliant areas and gray control blind areas automatically identified by the system, combined with the mine site construction specifications, coal seam geological conditions, and borehole construction process constraints, the intelligent borehole filling design unit of the three-dimensional visualization early warning and decision-making module 130 of this embodiment of the invention adopts a step-by-step refined design process to automatically generate a feasible extraction borehole filling optimization design scheme, including: dividing the continuous non-compliant and control blind area grid into several borehole filling treatment sub-regions according to geological structure, mining face division, and borehole construction accessibility; preliminarily screening the basic parameters of borehole filling based on the coal seam conditions and residual gas parameters of each treatment sub-region; optimizing the borehole position based on the spatial uniform coverage algorithm to ensure that the extraction influence range of adjacent boreholes meets the preset overlap rate requirements, and that the distance between the borehole position and existing roadways and constructed boreholes complies with construction safety specifications; outputting a borehole filling scheme containing borehole position coordinates, borehole parameters, and construction suggestions, and simulating and displaying the extraction coverage range after borehole filling in the GIS three-dimensional scene. The specific steps of the step-by-step refined design process are as follows:

[0095] (1) Refined division of early warning area: The continuous non-compliant / blind area grid is divided into several independent sub-areas for borehole repair. The division is based on the geological structure of the coal seam (faults, folds, etc.), the division of mining faces, and the accessibility of borehole construction. Each sub-area is an independent borehole repair design unit.

[0096] (2) Initial screening of borehole parameters: Based on the residual gas parameters, coal seam thickness, burial depth and other conditions of each treatment sub-area, and in conjunction with the "Design Specification for Coal Mine Gas Drainage Engineering", the basic parameters of the borehole are initially screened. The borehole diameter is adapted to the existing construction equipment in the mine, and 90mm / 113mm is commonly used. The borehole inclination angle is determined according to the coal seam occurrence angle to avoid crossing the layer. The borehole depth covers the entire treatment sub-area and does not exceed the mining planning range.

[0097] (3) Intelligent optimization of borehole location: Based on the spatial uniform coverage algorithm, the location of the supplementary borehole is optimized in the treatment sub-region. The arrangement of borehole locations meets two constraints: First, the overlap rate of the extraction influence range of adjacent boreholes is not less than 30% to avoid extraction gaps; Second, the distance between the borehole location and the existing roadway and the existing boreholes complies with the construction safety specifications, and the distance from the roadway wall is not less than 0.5m and the distance from the existing boreholes is not less than 2m.

[0098] (4) Output and verification of the hole repair scheme: The system automatically generates the hole repair optimization design scheme for each treatment sub-area. The scheme includes key parameters such as hole location coordinates, borehole inclination angle, depth, hole diameter, expected extraction time, and supporting construction equipment suggestions. At the same time, the extraction coverage area after hole repair is simulated and displayed in the GIS 3D scene. It supports manual adjustment of parameters and real-time verification of coverage effect. Finally, the output is a hole repair scheme drawing and parameter table that can be directly used for on-site construction. The format is compatible with the existing computer-aided design (CAD) / building information model (BIM) design software of the mine.

[0099] The edge-cloud collaborative computing architecture 140 includes an explosion-proof mining edge computing node deployed underground and a cloud computing platform deployed on the ground. The edge computing node is used to perform data acquisition, preprocessing and lightweight computing tasks, and the cloud computing platform is used to perform model training and prediction calculations and 3D visualization rendering.

[0100] S4: Building an Edge-Cloud Collaborative Computing Architecture 140

[0101] The system adopts a two-tier architecture of "underground mine explosion-proof edge nodes + ground cloud platform," clearly defining the functions and computing tasks of the edge and cloud. The hardware configuration of the underground edge nodes (including explosion-proof, computing power, storage, and communication) meets the explosion-proof and environmental adaptability requirements of coal mines. Specifically:

[0102] S41: Downhole edge node

[0103] (1) Hardware selection requirements: The intrinsically safe / explosion-proof edge computing equipment for mining shall be adopted and shall obtain the mining product safety mark (MA) certificate. The explosion-proof certification level shall be Ex ib I Mb (intrinsically safe type for mining) / Ex d I Mb (explosion-proof type for mining). The environmental adaptability shall meet the following requirements: working temperature -20℃~60℃, relative humidity 0~95% (non-condensing), dustproof and waterproof enclosure protection level (IP65) or above, and vibration resistance and electromagnetic interference resistance shall meet the requirements of the underground coal mine environment.

[0104] (2) Specific hardware configuration standards: The core computing capabilities required for underground edge nodes are as follows: the central processing unit (CPU) must have at least 4 cores and a main frequency of not less than 2.0 GHz; it must have a computing power unit that supports deep learning inference, and its computing power should not be less than 16 TOPS (for example, an NPU or a GPU with equivalent computing power can be used). As an example product that meets this standard, Huawei Atlas 500 Pro intrinsically safe intelligent station for mining or Advantech UNO-4683 explosion-proof terminal for mining can be selected. Fourth-generation double data rate synchronous dynamic random access memory (DDR4) ≥16G, solid-state drive (SSD) ≥512G (supports mining shockproof hard drive); the communication module is equipped with a mining gigabit Ethernet port (RJ45, compliant with MT / T 1004 standard) as standard, and optional mining 5G intrinsically safe module (supports NR-U underground dedicated frequency band), RS485 / 232 serial port (adapted to traditional underground monitoring equipment); other functions support local data caching (caching ≥7 days of data after network disconnection), remote wake-up, and fault self-diagnosis.

[0105] (3) Core functions: Deployed in underground mining faces, extraction pump stations and other field areas, it mainly completes lightweight calculations such as data acquisition, standardized preprocessing, outlier detection, kriging interpolation, basic association between extraction data and grid, and local data caching. At the same time, it realizes local storage and real-time uploading of field data, reducing the data transmission bandwidth requirements.

[0106] S42: Ground-to-Cloud Platform

[0107] (1) Hardware configuration: Cloud server cluster is adopted, and it is recommended to have more than 2 main and backup machines. The core configuration is: CPU ≥ 24 cores, DDR4 ≥ 128G, hard disk ≥ 4T independent disk redundant array level 5 (RAID5), GPU ≥ 3090 (or computing power card ≥ 32TOPS), and supports elastic expansion.

[0108] (2) Core functions: Deployed in the mine ground dispatch center, it mainly completes complex calculations such as deep learning model training and incremental update, high-precision inversion of residual gas parameters, fine design of hole filling scheme, GIS three-dimensional visualization rendering, long-term data storage and statistical analysis, and model distribution. At the same time, it receives data uploaded by edge nodes and issues control and configuration instructions. The cloud platform supports access from multiple terminals, including the dispatch center screen, duty computer, and mobile APP.

[0109] S43: Edge-Cloud Data Interaction

[0110] The edge and cloud are connected by a dual-mode network of mining gigabit industrial Ethernet + 5G intrinsically safe communication, which follows the MT / T1146 coal mine industrial Ethernet standard. Data transmission adopts a 256-bit advanced encryption standard (AES-256) encryption protocol to ensure data security. At the same time, it realizes the function of resuming interrupted transmission to avoid data loss due to network interruption.

[0111] Optionally, in some embodiments of the invention, the system further includes: a phased implementation strategy and a security authentication specification.

[0112] Step S5: Implement the system's phased strategy and security certification specifications

[0113] Step S51: Three-tiered phased implementation strategy

[0114] Step S5 is used to develop a three-tiered, phased implementation strategy to suit the needs of mines of different sizes.

[0115] The system adopts a modular and upgradeable design, allowing mines of different sizes to choose the appropriate version based on their needs and budget. Each version is compatible with standardized data interfaces, facilitating future expansion. Mines can gradually upgrade their functions according to their own development, avoiding high one-time costs. Specifically:

[0116] (1) Basic version (low-cost implementation version): The core functions are basic modeling of gas geology three-dimensional grid, grid attribute update driven by mining advance, standardized access of extraction data, basic judgment of extraction compliance status and GIS three-dimensional visualization early warning. It only uses Kriging interpolation for simple parameter deduction, without the need to deploy complex deep learning models. Edge nodes complete most of the calculations, and the cloud only performs data storage and visualization display, which is suitable for the basic needs of small and medium-sized mines.

[0117] (2) Advanced version (precision improvement version): Based on the basic version, it adds ST-GCN spatiotemporal prediction model, robust inversion of residual gas parameters, and intelligent initial screening function for hole filling design. The edge-cloud collaborative prediction calculation is completed. It is suitable for mines with a certain information infrastructure and can realize the forward-looking inference of gas parameters and preliminary intelligent hole filling design.

[0118] (3) Advanced version (full-function intelligent version): Based on the advanced version, after verification by small samples on site, it is upgraded to ST-GCSN high-precision prediction model, step-by-step fine hole filling design, and full-process closed-loop management to realize full-function intelligent "modeling-deduction-evaluation-early warning-design" and adapt to the high-end needs of large-scale modern mines.

[0119] S52: System Full-Process Security Authentication and Implementation Specifications

[0120] Step S52 is used to clarify the system's full-process security certification requirements and implementation specifications to ensure the safety and environmental adaptability of downhole deployments.

[0121] S521: Core Security Certification Requirements

[0122] The system as a whole complies with the "Coal Mine Safety Regulations" (2022 edition), the "Management Measures for Safety Marks of Mining Products", and the "General Technical Requirements for Coal Mine Safety Monitoring Systems"; all underground equipment must obtain the safety mark (MA) for mining products and the explosion-proof certificate; the network architecture adopts a physical isolation design, the underground control network and the ground management network are completely isolated, cross-network access is prohibited, and if the ground cloud platform needs to access the external network, a firewall and an intrusion detection system (IDS) must be added.

[0123] S522: Interface Implementation Specification

[0124] The system interfaces with existing underground monitoring and mining equipment using standard coal mining industry protocols, including Modbus-RTU, Unified Architecture Open Platform Communication (OPC UA), MT / T 1004, etc. If the equipment uses a non-standard protocol, it is necessary to coordinate with the equipment manufacturer to complete the development and adaptation of customized interfaces, and conduct interface stability testing before putting it into use.

[0125] S523: On-site Implementation Specifications

[0126] The implementation of data standardization and specifications is accompanied by on-site operator training in mines. The training content includes data acquisition, anomaly handling, and equipment operation. Only those who pass the training and assessment can be allowed to work. The underground installation of edge nodes should follow the coal mine underground equipment installation specifications, be kept away from water-spraying and roof-falling areas, and take good anti-vibration and dust prevention measures.

[0127] Optionally, in some embodiments of the invention, the system constructs a closed-loop management mechanism covering the entire process of "modeling-deduction-evaluation-early warning-design-construction-feedback," specifically including the following steps:

[0128] S6: System Overall Closed-Loop Workflow

[0129] This invention constructs a modular, implementable, closed-loop, and highly secure intelligent management cycle, balancing algorithm robustness and engineering practicality while preserving the potential for further optimization of the ST-GCSN model. The overall system workflow is as follows:

[0130] (1) In the system initialization phase, a three-dimensional meshed basic model of coal seam gas geology was constructed based on the actual situation of the mine and initial attributes were assigned. The computational task division of the edge-cloud collaborative architecture was configured, and the installation and debugging of underground edge nodes were completed, with the installation and debugging complying with explosion-proof and safety specifications. At the same time, data standardization and equipment operation training were conducted for on-site personnel to ensure that the specifications were implemented. Finally, the construction of the three-dimensional voxel mesh of the coal seam, the deployment of the edge-cloud collaborative architecture, and the training on on-site data acquisition specifications were completed.

[0131] (2) During the data acquisition and model simulation stage, the underground edge nodes automatically acquire mining footage data daily, and simultaneously collect, preprocess, and cache extraction monitoring data locally; and complete the activation and standardized attribute update of the exposed grid according to the core data quality threshold in step S122.

[0132] (3) In the stage of spatiotemporal extrapolation and dynamic model correction of gas geological parameters, edge nodes complete the preliminary extrapolation of Kriging interpolation, and upload the feature data to the cloud after encryption by AES-256. The cloud uses the ST-GCN / ST-GCSN model to complete the high-precision spatiotemporal prediction of gas geological parameters in unexposed areas according to the system version deployed in the mine. The final gas geological parameter extrapolation results are obtained by fusion, and the dynamic correction of the three-dimensional gridded basic model of coal seam gas geology is completed.

[0133] (4) In the intelligent evaluation stage of extraction compliance, the cloud receives the standardized extraction data uploaded by the edge node, and calculates the residual gas parameters of each grid according to the robust inversion formula in step S22 through the intelligent calculation engine of extraction compliance, and automatically completes the extraction compliance status judgment of each grid unit according to the four-level fault-tolerant judgment rule in step S23.

[0134] (5) In the stage of visualization early warning and intelligent decision-making, the cloud maps the results of the extraction compliance status judgment to the three-dimensional grid basic model of coal seam gas geology, dynamically generates and updates the four-color early warning chromatogram cloud map for visualization display, realizes visualization monitoring and hidden danger early warning of extraction compliance status, and the mine ground dispatch center can view and issue treatment instructions in real time; for non-compliant areas and control blind spots, the cloud automatically generates extraction and hole filling optimization design schemes that conform to the on-site construction specifications according to the step-by-step refined process of step S32, supports manual adjustment and outputs construction drawings and parameter tables to guide on-site construction.

[0135] (6) Closed-loop feedback and model update stage: new data generated after the on-site construction according to the hole filling plan is collected and uploaded through edge nodes. The edge nodes collect new extraction data and gas measurement data in real time, and upload them to the cloud after standardization according to the specifications of step S122. The cloud performs ST-GCN / ST-GCSN model fine-tuning according to the incremental learning mechanism based on the new data to ensure the model prediction accuracy. The full data is regularly integrated for model retraining to realize the continuous optimization of the gas geological model and the dynamic update of the extraction standard status, forming a closed-loop management of the whole process of "modeling-deduction-evaluation-early warning-design-construction-feedback".

[0136] (7) Based on its own development and the effect of field application, the mine gradually upgrades the system function modules, iterating from the basic version to the advanced version and the high-level version; in the high-level version, the ST-GCSN model is retrained every 3 months according to the regular retraining mechanism to continuously optimize the prediction effect.

[0137] Through the coordinated operation of the above stages, the system achieves fully automated closed-loop management of the entire process, from gas geological modeling, parameter extrapolation, compliance assessment, hazard warning, treatment design to on-site construction feedback, enabling gas control to transform from passive response to proactive prevention and control.

[0138] In summary, this invention, through steps S1-S6, deeply integrates Geographic Information System (GIS), dynamic gridded modeling, phased spatiotemporal map algorithm, robust inversion evaluation engine, and edge-cloud collaborative architecture. It clarifies all core algorithm parameters, hardware configuration standards, data quality thresholds, and security certification requirements, achieving integrated, transparent, and forward-looking intelligent management of mine gas geology and extraction processes. Simultaneously, through phased implementation, modular design, and edge computing architecture, it significantly reduces system deployment and maintenance costs, adapting to the needs of mines of different sizes. Implementation standards, including on-site training, vendor interface coordination, and on-site model verification, mitigate the risks associated with idealized technology implementation. This invention effectively solves the problems of lag in static evaluation and blind spots in manual management in traditional technologies, significantly improving the refinement, intelligence, and safety assurance capabilities of coal mine gas management. All technical solutions meet the safety and environmental compatibility requirements of underground coal mines.

[0139] Table 1

[0140]

[0141] Figure 5 Table 1 shows the simulation results comparing the loss values ​​of ST-GCSN and ST-GCN under training / test data in this invention; Table 2 shows the performance comparison results of the ST-GCSN and ST-GCN algorithms in this invention. From Table 1 and... Figure 5Simulation results show that ST-GCSN exhibits significant performance advantages over ST-GCN: In terms of training efficiency, ST-GCSN's training time is only 2.5 hours, far less than ST-GCN's 10 hours, and the model parameter size is only 0.23M. The single-step test time also decreased from 2.95 milliseconds to 0.72 milliseconds. In terms of the trend of loss value, as the number of iterations increases, the loss value of ST-GCSN under both training and test data is consistently lower than that of ST-GCN. After 1000 iterations, the training loss value of ST-GCSN is 0.191, far lower than ST-GCN's 0.332, and the test loss value also remains at a lower level. This indicates that ST-GCSN has better fitting effect and generalization ability while ensuring lightweight and high inference speed.

[0142] It should also be noted that, in the embodiments of this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, 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 one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0143] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in the embodiments of this application may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown in this application, but is to be accorded the widest scope consistent with the principles and novel features disclosed in the embodiments of this application.

Claims

1. A GIS three-dimensional visualization mine gas geology dynamic evolution and extraction standard reaching intelligent evaluation system, characterized in that, include: The 3D dynamic modeling module is used to discretize the target coal seam into a 3D voxel grid and construct a basic gas geological model. It dynamically updates the properties of the exposed grid based on mining progress data and performs spatiotemporal extrapolation of the gas geological parameters of the unexposed area to achieve dynamic correction of the model. The three-dimensional dynamic modeling module includes: a basic model construction unit, used to discretize the target coal seam area into a three-dimensional voxel grid and assign initial gas geological attribute parameters to each grid; a grid dynamic activation and update unit, used to interface with the mine mining progress measurement equipment, automatically acquire mining spatial coordinate data, identify and activate the exposed grid cells based on the spatial coordinate data, and associate the on-site measured gas geological data with the corresponding activated grid cells after standardization processing to complete the attribute update of the exposed grids; and a spatiotemporal extrapolation unit, used to perform fusion calculations on the unexposed grid cells using a spatial interpolation algorithm and a spatiotemporal graph neural network algorithm, combined with the measured data of the exposed grids, to obtain predicted values ​​of gas geological parameters in the unexposed area, and dynamically correct the three-dimensional grid model based on the predicted values. The spatiotemporal extrapolation unit employs a phased iterative strategy. The first phase uses a spatiotemporal graph convolutional network for prediction, while the second phase, after verification with field data, upgrades to a spatiotemporal graph complementary scattering network for prediction. The spatiotemporal extrapolation unit weights and fuses the spatial interpolation algorithm's calculation results with the spatiotemporal graph neural network algorithm's prediction results. The fusion weights are dynamically adjusted based on historical prediction errors to obtain the final predicted values ​​of gas geological parameters in the unexposed area. The spatial interpolation algorithm uses the Kriging interpolation algorithm, and its calculation task is performed at the downhole edge computing node. The prediction calculation task of the spatiotemporal graph neural network algorithm is performed on a cloud computing platform. The intelligent evaluation module for gas extraction compliance is used to acquire and standardize gas extraction data, establish a spatial relationship between the standardized extraction data and the three-dimensional voxel grid, calculate the residual gas parameters of each grid in real time through the built-in inversion algorithm, and automatically determine the extraction compliance status of each grid according to preset rules. The 3D visualization early warning and decision-making module is used to map the judgment results of the extraction compliance status to a 3D voxel grid in a visual form and display them dynamically. It automatically generates extraction and hole filling treatment plans for areas judged as non-compliant. The edge-cloud collaborative computing architecture includes explosion-proof mining edge computing nodes deployed underground and cloud computing platforms deployed on the ground. The edge computing nodes are used to perform data acquisition, preprocessing, and lightweight computing tasks, while the cloud computing platform is used to perform model training and prediction calculations and 3D visualization rendering.

2. The GIS three-dimensional visualization mine gas geology dynamic evolution and extraction standard attainment intelligent evaluation system according to claim 1, characterized in that, The spatiotemporal graph complementary scattering network uses graph complementary scattering layers as basic modules. Each graph complementary scattering layer contains fixed feature extraction branches and trainable complementary feature extraction branches. The fixed feature extraction branch is generated by spatiotemporal graph scattering transform optimized by pruning technology, and spatiotemporal features of gas geological parameters are extracted by mathematically designed spatial wavelet filters and temporal wavelet filters. The trainable complementary feature extraction branch adopts a filter structure that is complementary to the fixed feature extraction branch, including a trainable spatial complementary filter and a temporal complementary filter. A trainable graph shift matrix is ​​generated through a surrogate parameter training mechanism. The surrogate parameter training mechanism is to obtain a trainable matrix that conforms to the characteristics of a Markov matrix after the surrogate parameter matrix is ​​row-normalized. The spatiotemporal complementary scattering network is trained using a hybrid optimization strategy. An adaptive moment estimation optimizer is used in the early stage, and a stochastic gradient descent optimizer is used for fine-tuning in the later stage. The loss function is a combination of mean squared error and L1 regularization. The spatiotemporal graph complementary scattering network adopts an online update mechanism that combines incremental learning and periodic retraining. The incremental learning cycle is a preset number of days, during which small-batch fine-tuning is performed on the original model. The periodic retraining cycle is a preset number of months, during which the model is fully retrained by integrating all newly added data.

3. The GIS-based three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance as described in claim 1, characterized in that, The intelligent evaluation module for sampling compliance includes: The extraction data access and processing unit is used to interface with the downhole gas extraction monitoring system, automatically acquire data on extraction flow rate, gas concentration and extraction duration of the extraction borehole, and perform outlier filtering and missing value completion processing on the acquired data according to the preset data quality threshold. The grid-bore intelligent association unit is used to establish the association relationship between each grid unit and the extraction boreholes within its influence range based on the spatial topological relationship between the borehole spatial coordinates and the three-dimensional voxel grid, and to determine the extraction influence weight of each borehole on the grid according to the spatial distance between the borehole and the grid. The residual gas parameter inversion calculation unit is used to calculate the residual gas content and residual gas pressure of each grid cell in real time using a built-in inversion algorithm, based on the original gas content and original gas pressure of each grid cell, as well as the cumulative drainage flow rate, average gas concentration, and drainage duration of the associated drainage boreholes, combined with the grid coal seam volume, apparent coal density, and gas drainage utilization coefficient. The inversion algorithm introduces a data fault tolerance mechanism, which uses the effective drainage data of surrounding grids for collaborative calculation when the drainage data of a single grid is incomplete. The automatic compliance status determination unit is used to automatically compare the residual gas content and residual gas pressure obtained from the inversion calculation with the extraction compliance threshold value specified in the industry standard. According to the preset four-level fault-tolerant determination rule, each grid unit is determined to be one of the four states: compliant, critical warning, non-compliant, or control blind zone.

4. The GIS three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance as described in claim 3, characterized in that, The four-level fault-tolerant judgment rule adopted by the automatic qualification status determination unit is as follows: When the residual gas parameter is below 80% of the critical value for drainage compliance, and the associated drainage data is complete and valid, the grid cell is determined to be in compliance status. When the residual gas parameter is between 80% and 100% of the critical value for drainage compliance, and the associated drainage data is complete and valid, the grid cell is determined to be in a critical warning state. When the residual gas parameter is higher than the critical value for achieving the extraction standard, and the associated extraction data is complete and valid, the grid cell is determined to be in a non-compliant state. When a grid cell has no effective extraction borehole coverage, or the associated extraction data and measured data are abnormally missing and cannot be filled in, or the inversion calculation results exceed the preset reasonable range, the grid cell is determined to be in a control blind zone state.

5. The GIS three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance as described in claim 1, characterized in that, The 3D visualization early warning and decision-making module includes: The four-color early warning visualization unit is used to map the results of the compliance status judgment of each grid output by the intelligent evaluation module for sampling compliance to the three-dimensional voxel grid in real time. In the GIS three-dimensional scene, the four colors of green, yellow, red and gray correspond to the compliance, critical warning, non-compliance and control blind zone status respectively, forming a dynamically updated four-color early warning color spectrum cloud map. The intelligent borehole filling design unit is used to automatically generate optimized design schemes for extraction borehole filling in continuous grid areas that are identified as substandard or control blind zones, taking into account mine construction specifications and geological conditions.

6. The GIS-based three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance as described in claim 5, is characterized in that... The four-color early warning visualization unit adopts an edge-cloud collaborative update mechanism, in which data changes are pushed in real time by the downhole edge node and lightweight rendering is performed in the cloud to achieve second-level dynamic refresh of the color cloud map.

7. The GIS three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance as described in claim 5, characterized in that, The intelligent hole-filling design unit adopts a step-by-step refined design process, including: The continuous non-compliant and control blind area grids are divided into several sub-regions for borehole repair and treatment according to geological structure, mining face division and accessibility of drilling construction; Based on the coal seam conditions and residual gas parameters of each treatment sub-region, the basic parameters for borehole filling were initially selected. The hole locations are optimized based on the spatial uniform coverage algorithm, so that the extraction influence range of adjacent boreholes meets the preset overlap rate requirements, and the distance between the hole locations and existing roadways and existing boreholes complies with construction safety specifications. The output includes borehole location coordinates, drilling parameters, and construction suggestions for the borehole repair scheme, and simulates and displays the extraction coverage area after borehole repair in a GIS 3D scene.

8. The GIS three-dimensional visualization intelligent evaluation system for dynamic evolution of mine gas geology and extraction compliance as described in claim 6, characterized in that, The edge-cloud collaborative update mechanism adopted by the four-color early warning visualization unit includes: The underground edge nodes only push the extraction compliance status data of the grid cells that have changed. The cloud computing platform locally refreshes the rendering attributes of the corresponding grid cells in the 3D scene based on the incremental update algorithm, without reloading and rendering the entire 3D scene. The cloud computing platform has a built-in GIS offline rendering engine, which supports the continuous provision of 3D visualization viewing function based on the last synchronized full status data in the event of a mine network interruption. After the network is restored, it automatically synchronizes data and completes status with the edge nodes.

Images