A method and system for early warning of dynamic disasters in coal mines based on stress field analysis and multi-data source fusion

By constructing an initial geostress field model and fusing multi-source data using a graph neural network, the problem of insufficient data fusion in coal mine dynamic disaster early warning was solved, enabling accurate analysis and intelligent early warning of dynamic stress fields, and improving the timeliness and accuracy of early warning.

CN123119620APending Publication Date: 2026-11-17PINGDINGSHAN TIANAN COAL MINING +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611028617.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-10
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies cannot achieve deep integration of multi-source data, making it difficult to fully reflect changes in geological conditions and mechanical states in coal mine dynamic disaster early warning, and failing to provide accurate stress field data support, resulting in a prominent problem of delayed disaster early warning.

Method used

By acquiring mine geological parameters, coal and rock mechanical parameters, and measured underground in-situ stress parameters, an initial in-situ stress field model is constructed. Combined with a multi-dimensional dynamic disaster sensitivity index evaluation system, a graph neural network is used for data fusion and dynamic correction to generate a dynamic stress field model and output the disaster early warning level.

Benefits of technology

It enables timely and accurate early warning of dynamic disasters in coal mines, improves the comprehensiveness and reliability of early warning, reduces the false alarm rate, ensures data flow continuity and closed loop, and supports intelligent optimization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123119620A_ABST
    Figure CN123119620A_ABST
Patent Text Reader

Abstract

This invention relates to the field of coal mine safety detection and early warning technology, and provides a method and system for early warning of dynamic disasters in coal mines based on stress field analysis and multi-data source fusion. The method includes: first, integrating geological, mechanical, and measured parameters to establish and verify an initial geostress field; then, constructing a sensitive index system including stress state, geological structure, coal and rock characteristics, and gas-related indicators, and dynamically determining index weights using the analytic hierarchy process (AHP) and a data-driven model; next, using detection equipment to collect data in real time, and dynamically correcting the initial stress field through data fusion and adaptive algorithms; finally, constructing a large-scale model multi-data fusion system, combining graph neural networks and a dynamic weighting mechanism to accurately analyze and determine early warning levels for different mining scenarios. This invention solves the problems of single data source, insufficient fusion, and delayed early warning in traditional methods, achieving intelligent, dynamic, and accurate early warning of dynamic disasters in deep coal mines.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of coal mine safety detection and early warning technology, and in particular to a method and system for early warning of dynamic disasters in coal mines based on stress field analysis and multi-data source fusion. Background Technology

[0002] Accurately establishing the geostress field is a crucial foundation for early warning of dynamic disasters in coal mines. Traditional methods for establishing geostress fields have significant shortcomings in integrating multi-source parameters: at the level of mine geological data acquisition, the means of obtaining data on the thickness and occurrence of strata lithology (such as coal seams, sandstone, mudstone, etc.) and the distribution of geological structures (faults, folds, etc.) are limited, making it difficult to achieve high-precision mapping. Taking the acquisition of geological structure parameters as an example, parameters such as the curvature of synclines and anticlines, the precise location of faults (normal faults, reverse faults), and parameters on coal seam thickness variations rely on manual surveys and a limited number of borehole samples. This results in low data density and significant errors, making it impossible to accurately simulate the impact of geological structures on the stress field.

[0003] Regarding the acquisition of measured parameters, traditional methods such as hydraulic fracturing, stress relief, and spherical stress sensors can obtain some data, but they suffer from problems such as incomplete data and accuracy being affected by various factors. For example, hydraulic fracturing is affected by the degree of rock mass fracture development, resulting in large data dispersion; stress relief is complex to operate, requires high technical skills from construction personnel, and is prone to human error. These measured data are difficult to effectively couple with tectonic stress and stress fields to establish accurate mathematical models. Furthermore, when comparing and verifying measured data with numerical simulation methods, the error correction effect is poor due to data limitations, failing to provide accurate stress field data support for dynamic disaster early warning.

[0004] In the field of evaluating sensitive indicators for dynamic hazards, existing evaluation systems struggle to comprehensively and accurately reflect the complex geological conditions and changes in mechanical state during coal mining. Different mining areas, such as tunneling faces and coal mining faces, exhibit significantly different mining characteristics and risk levels, but existing evaluation systems fail to adequately consider these differences when assigning reasonable data weights. In tunneling faces, the stress concentration caused by rapid tunnel advancement is a significant contributing factor to dynamic hazards; therefore, greater attention should be paid to data on stress changes. However, existing evaluation systems often employ average weight allocation, weakening the early warning value of key data.

[0005] For coal mining faces, stress data of the working face and roof and floor under the influence of mining are crucial for disaster early warning. However, the existing evaluation system also lacks targeted weight adjustment, resulting in insufficient accuracy in the evaluation of sensitive indicators of dynamic disasters. This makes it impossible to provide a reliable basis for disaster early warning, and the problem of delayed disaster early warning is prominent in practical applications.

[0006] Traditional methods for correcting and analyzing dynamic stress fields primarily rely on data acquired by spherical stress detectors and microseismic detectors. However, these data sources are relatively singular and cannot comprehensively cover the complex changes in the stress field during coal mining. In data processing and analysis, traditional methods lack effective integration and dynamic correction mechanisms for the initial stress field, making it impossible to track the dynamic evolution of the stress field under mining disturbances in real time. Due to the complex and variable coal mining environment, factors such as stress redistribution and geological structure activation caused by mining activities mean that a single data source cannot fully reflect the dynamic changes in the stress field. This leads to inaccurate analysis of the dynamic stress field, making it difficult to detect abnormal changes in the stress field before dynamic disasters occur, thus missing the optimal early warning opportunity.

[0007] Current technologies have significant shortcomings in multi-data fusion and large-scale model analysis and early warning. Existing technologies cannot achieve deep fusion of multi-source data such as geostress field data, dynamic disaster sensitive index data, microseismic monitoring data, geological structure data, and coal (rock) mechanical property data. There is a lack of effective multi-data fusion mathematical models, making it impossible to set scientific and reasonable weight adjustment strategies based on the importance and correlation of different data, and thus failing to comprehensively reflect changes in geological conditions and mechanical states during coal mining.

[0008] In summary, existing technologies have many shortcomings in early warning of coal mine dynamic disasters and cannot meet the needs of safe and efficient coal mining. Therefore, there is an urgent need for a dynamic disaster early warning method that can achieve multi-data fusion, accurately predict sensitive indicators of mine dynamic disasters, and perform precise 24-hour analysis based on a large-scale model, in order to improve the level of coal mine dynamic disaster prevention and control and ensure the sustainable development of the coal industry. Summary of the Invention

[0009] To address the problem that existing technologies cannot deeply integrate multi-source data, making it difficult to fully reflect changes in geological conditions and mechanical states during coal mining, this invention proposes a large-scale dynamic disaster early warning method and system based on stress field analysis and multi-source data fusion.

[0010] In a first aspect, the present invention provides a method for early warning of dynamic disasters in coal mines based on stress field analysis and multi-data source fusion, comprising:

[0011] Step 1: Acquire and integrate multi-source data including mine geological parameters, coal and rock mechanics parameters, and measured underground in-situ stress parameters, and construct an initial in-situ stress field model using numerical simulation.

[0012] Step 2: Based on the multi-source data and the initial geostress field model, establish a dynamic disaster sensitivity index evaluation system that includes four dimensions of indicators: stress state, geological structure, coal and rock characteristics, and gas-related indicators. Use the analytic hierarchy process (AHP) to determine the weight of each indicator and construct a comprehensive evaluation function for risk classification.

[0013] Step 3: By acquiring real-time data from detection equipment deployed at the tunneling face, coal mining face, and mining area, the real-time data is compared and analyzed with the initial geostress field model to construct a difference function to determine the dynamic correction coefficient. Based on the dynamic correction coefficient of each region, the initial geostress field model is dynamically corrected to generate a dynamic stress field model.

[0014] Step 4: Construct a graph neural network fusion model. The dynamic stress field model, risk classification results, and real-time data are used as inputs to the graph neural network fusion model. The input data is fused using a message passing mechanism, and upper-level comprehensive dynamic weights are introduced to perform weighted optimization on the fused data to obtain the final fused features. The final fused features are then input into a preset early warning level determination model to output the disaster early warning level.

[0015] Furthermore, step 1 specifically includes:

[0016] Step 1.1: Obtain mine geological parameters and coal and rock mechanics parameters, specifically including: obtaining data on the thickness and occurrence of strata lithology; obtaining the curvature radius parameters of synclines and anticlines of fold structures; obtaining the strike, dip angle, length, and displacement parameters of faults, and calculating fault activity rate parameters based on microseismic monitoring data; and constructing normal fault fusion models and reverse fault fusion models based on the fault activity rate parameters to correct the geostress.

[0017] Step 1.2: Obtain measured parameters of downhole in-situ stress, including: measured data of triaxial principal stress, stress gradient and six-axis stress components of rock mass obtained by hydraulic fracturing method or stress relief method;

[0018] Step 1.3: Obtain the correlation characteristic data of geostress direction, stress anomaly and tectonic stress energy, including: microseismic event location parameters and source energy parameters obtained from microseismic monitoring data inversion, and seismic wave velocity anomaly parameters and reflected wave imaging parameters obtained from 3D seismic exploration data inversion;

[0019] Step 1.4: Using the mine geological parameters, coal and rock mechanical parameters, and tectonic stress energy correlation characteristic data as input, perform simulation calculations using finite element or discrete element numerical models to generate simulation results of the global initial geostress field;

[0020] Step 1.5: Compare and verify the simulation results of the global initial geostress field with the measured parameters of the downhole geostress field and iteratively correct them until the simulation result error is lower than the preset threshold, and generate the initial geostress field model.

[0021] Furthermore, in step 2, the stress state indicators include the maximum principal stress concentration factor, stress gradient, and stress-gas coupling hazard index; the geological structure indicators include the fault influence factor and fold curvature index; the coal and rock properties indicators include the coal impact tendency index and rock brittleness index; and the gas-related indicators include gas pressure, gas content, gas dissipation index, borehole gas emission initial velocity, drill cuttings volume, and rock damage evolution coefficient.

[0022] Furthermore, in step 3, the formula for the difference function is:

[0023]

[0024] in, This represents the stress change vector between real-time data and the initial geostress field model. Assigning weights to stress data in each direction. This represents the stress vector of the initial geostress field model. This represents the geostress field vector of real-time data.

[0025] Based on the aforementioned difference function, the stress field is updated using an adaptive correction algorithm, and the correction formula is as follows:

[0026]

[0027] in, This represents the stress vector in a long-range dynamic stress model. This represents the correction factor. This represents the stress variation vector.

[0028] Furthermore, step 4, the construction of the graph neural network fusion model, specifically includes:

[0029] Using stress, microseismic, and gas monitoring points of underground mine monitoring equipment as nodes of a graph neural network, the spatial distance between nodes is calculated based on the spatial coordinates of the monitoring points, and the edges of the graph are constructed based on the geological correlation between the monitoring points.

[0030] Calculate the feature vector of each node; wherein the features of the stress node are calculated using the dynamic stress field model data, the features of the microseismic node are calculated using real-time microseismic monitoring data, the features of the gas node are calculated using the gas monitoring data, and the features of each node are integrated with the index data in the sensitive index evaluation system.

[0031] Based on the edge weights, each node aggregates the feature information of its neighboring nodes through the message passing mechanism of the graph neural network, and updates the node features to complete the input feature fusion:

[0032]

[0033] in, This represents the feature vector of node i after the (t+1)th round of message passing; This represents the feature vector of node i after the t-th round of message passing; This represents the edge weight between node i and node j; This represents the feature vector of node j after the t-th round of message passing; Represents the set of neighboring nodes of node i; Indicates the activation function;

[0034] A second weighting of the fused features is applied by introducing upper-level comprehensive dynamic weights to obtain the final fused features.

[0035] Furthermore, the upper-layer integrated dynamic weights, after the graph neural network fusion model outputs node fusion features, perform secondary weighted optimization on the output features, specifically including:

[0036]

[0037] in, This represents the overall weight of the k-th type of node. Represents the balance coefficient. This represents the objective weights calculated using the entropy weight method. This represents the subjective weights determined using the analytic hierarchy process (AHP).

[0038] Furthermore, the process of inputting the fused features into the early warning level determination model to obtain the dynamic disaster early warning result specifically includes:

[0039] The features output by the graph neural network fusion model are mapped to a set of early warning judgment indicators;

[0040] Establish the membership function of each indicator in the early warning judgment indicator set to different early warning levels, and form a membership matrix;

[0041] The dynamic weight vector of the index is combined with the membership matrix by a fuzzy synthesis operation to obtain the comprehensive membership vector.

[0042] Based on the principle of maximum membership, the risk level of dynamic disasters is determined from the comprehensive membership vector, and a dynamic disaster early warning signal is output.

[0043] Furthermore, step 4 also includes: introducing a reinforcement learning real-time self-optimization mechanism, taking the monitoring point as the state of the agent, taking the parameters of the graph neural network fusion model as the action of the agent, taking the comprehensive index of warning accuracy and false alarm rate as the reward function, and adjusting the parameters of the graph neural network fusion model iteratively through the agent to achieve adaptive optimization of the graph neural network fusion model.

[0044] Secondly, this invention provides a coal mine large-scale model dynamic disaster early warning system based on stress field analysis and multi-data source fusion, comprising:

[0045] The initial modeling module is used to acquire and integrate multi-source data including mine geological parameters, coal and rock mechanics parameters, and measured underground geostress parameters, and to construct an initial geostress field model through numerical simulation.

[0046] The indicator evaluation module is used to establish a dynamic disaster sensitivity indicator evaluation system based on the multi-source data and the initial geostress field model. This system includes four dimensions of indicators: stress state, geological structure, coal and rock characteristics, and gas-related indicators. The weight of each indicator is determined by the analytic hierarchy process (AHP), and a comprehensive evaluation function is constructed for risk classification.

[0047] The dynamic correction module is used to acquire real-time data from detection equipment deployed at the tunneling face, coal mining face, and mining area, compare and analyze the real-time data with the initial geostress field model, construct a difference function to determine the dynamic correction coefficient, and dynamically correct the initial geostress field model according to the dynamic correction coefficient of each region to generate a dynamic stress field model.

[0048] The early warning output module is used to construct a graph neural network fusion model. The dynamic stress field model, risk classification results, and real-time data are used as inputs to the graph neural network fusion model. The input data is fused using a message passing mechanism, and upper-level comprehensive dynamic weights are introduced to perform weighted optimization on the fused data to obtain the final fused features. The final fused features are then input into a preset early warning level determination model to output the disaster early warning level.

[0049] The beneficial effects of this invention are as follows:

[0050] The dynamic disaster early warning method provided by this invention overcomes the limitations of single data sources by fusing dynamic stress fields, multi-dimensional sensitive indicators, real-time microseismic data, and gas data in a spatial topological structure using a graph neural network, resulting in more comprehensive early warning data. Furthermore, it ensures a coherent and closed-loop data flow from basic geological data to the final early warning signal, solving the data fragmentation problem of traditional methods. By introducing a reinforcement learning mechanism, the model can automatically optimize key parameters based on early warning feedback, continuously improving early warning accuracy and reducing false alarm rates, achieving long-term evolution in intelligence. Attached Figure Description

[0051] Figure 1 A flowchart illustrating a method for early warning of dynamic disasters in a large coal mine model based on stress field analysis and multi-data source fusion, provided as an embodiment of the present invention;

[0052] Figure 2 This is a structural diagram of a large-scale dynamic disaster early warning system for coal mines based on stress field analysis and multi-data source fusion, provided for an embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0054] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for early warning of dynamic disasters in coal mines based on stress field analysis and multi-data source fusion, comprising:

[0055] Step 1: Acquire and integrate multi-source data including mine geological parameters, coal and rock mechanics parameters, and measured underground in-situ stress parameters, and construct an initial in-situ stress field model through numerical simulation.

[0056] Specifically, the purpose of this step is to construct an initial geostress field model for coal mines. Its core is to integrate multiple sources of parameters such as coal mine geology, coal rock mechanics, and geostress parameters, establish an evaluation mathematical model that integrates geological structural characteristics and multi-physics field coupling, and finally obtain a high-precision initial geostress field model through iterative correction by numerical simulation and experimental verification.

[0057] Step 2: Based on multi-source data and the initial geostress field model, establish a dynamic disaster sensitivity index evaluation system that includes four dimensions: stress state, geological structure, coal and rock characteristics, and gas-related indicators. Use the analytic hierarchy process (AHP) to determine the weight of each indicator and construct a comprehensive evaluation function for risk classification.

[0058] Specifically, based on multi-source data and an initial geostress field model, an evaluation system for dynamic disaster sensitivity indicators is established, encompassing four dimensions: stress state, geological structure, coal and rock characteristics, and gas characteristics. In terms of weight determination, the analytic hierarchy process (AHP) is used to obtain the comprehensive basic weights of the indicators, and combined with dynamic parameter adjustments based on the working conditions of the mining face, a dynamic weight system adapted to the underground scenario is generated.

[0059] Step 3: By acquiring real-time data from detection equipment deployed at the tunneling face, coal mining face, and mining area, the real-time data is compared and analyzed with the initial geostress field model. A difference function is constructed to determine the dynamic correction coefficient. The initial geostress field model is dynamically corrected according to the dynamic correction coefficient of each area to generate a dynamic stress field model.

[0060] Specifically, by deploying advanced trenching, ground-penetrating radar, and microseismic equipment at tunneling faces, coal mining faces, and mining areas, real-time multi-source dynamic monitoring data is collected underground. Combined with the mine roadway layout and mining engineering layout, numerical simulation of excavation disturbance is carried out to recreate the stress redistribution effect of surrounding rock caused by roadway excavation and face advancement. A stress difference function is constructed to quantify the deviation between the real-time stress of mining disturbance and the initial geostress field. Combined with tectonic stress response and mining disturbance damage characteristics, the initial geostress field model is dynamically iteratively corrected to generate a dynamic stress field model that fits the roadway layout and real-time mining conditions.

[0061] Step 4: Construct a graph neural network fusion model. The dynamic stress field model, risk classification results, and real-time data are used as inputs to the graph neural network fusion model. The input data is fused using a message passing mechanism, and upper-level comprehensive dynamic weights are introduced to perform weighted optimization on the fused data to obtain the final fused features. The final fused features are then input into a preset early warning level determination model to output the disaster early warning level.

[0062] Specifically, a graph neural network fusion model with dual-layer weighted collaborative integration is constructed. Dynamic stress field model data, non-sensitive index evaluation system data, and downhole real-time monitoring data are used as inputs to the graph neural network fusion model. The spatial and geotechnical features of the measurement points are fused by relying on the message passing mechanism of the graph neural network fusion model. Secondary feature weighting optimization is completed by combining dynamic weights of working conditions. The final fused features are input into the early warning level determination model, and the dynamic disaster early warning result is output through fuzzy membership degree synthesis operation.

[0063] The method provided by this invention achieves a leap from static assessment to dynamic and accurate early warning of coal mine dynamic disasters by constructing a closed-loop process of "initial geostress field modeling - dynamic data-driven correction - multi-source intelligent fusion early warning", which significantly improves the timeliness, accuracy and reliability of early warning.

[0064] In one embodiment, step 1 specifically includes:

[0065] Step 1.1: Obtain mine geological parameters and coal and rock mechanics parameters. Specifically, this includes: obtaining data on the thickness and occurrence of strata lithology; obtaining the curvature radius parameters of synclines and anticlines of fold structures; obtaining the strike, dip angle, length, and displacement parameters of faults, and calculating fault activity rate parameters based on microseismic monitoring data. Based on the fault activity rate parameters, construct normal fault fusion models and reverse fault fusion models to correct for in-situ stress.

[0066] Specifically, geological parameters such as mine lithology, faults, folds, and coal seam occurrence are collected, as well as coal and rock mechanical parameters such as elastic modulus, Poisson's ratio, compressive strength, and brittleness index. In this embodiment, this step specifically includes:

[0067] (1) Basic geological and mechanical parameters: Through geological drilling and logging, the thickness and occurrence data of strata lithology (coal seam, sand layer, mudstone, gas-bearing strata, etc.) are analyzed and obtained. At the same time, through rock mechanics experiments, mechanical parameters such as coal and rock mass density, elastic modulus, Poisson's ratio, compressive strength, shear strength, cohesion, and internal friction angle are measured and a database is established. The above parameters are the basis for constructing the geostress field model, which is used to reflect the physical and mechanical properties and geological structure characteristics of coal and rock mass.

[0068] (2) Fold structure parameters: The curvature of synclines and anticlines is obtained by geological mapping and high-density three-dimensional seismic exploration data and by using geological structure analysis methods.

[0069] Specifically, the formula for calculating the radius of curvature R of synclines and anticlines is as follows:

[0070] ;

[0071] in, This represents the coordinates of the structural surface in the vertical direction. This represents the horizontal coordinate. A smaller radius of curvature R indicates more severe folding deformation and a higher degree of stress concentration in the corresponding region. Parameters such as the axial direction and the helix angle of the syncline and anticline are also recorded for subsequent stress field analysis.

[0072] (3) Geometric and activity parameters of faults: Geometric parameters such as strike, dip, dip angle, length, and displacement of faults are obtained through borehole data, seismic data, and fault information revealed by mine roadways. Among them, borehole data can provide information on the vertical position and displacement of faults; seismic data can present the spatial distribution of faults with a large detection range, and the strike, dip, and dip angle of faults can be determined relatively accurately through the analysis of seismic wave reflection signals; fault information revealed by mine roadways can intuitively obtain the specific characteristics of faults in the actual mining area, such as the width of the fault fracture zone.

[0073] Based on the relative movement direction of the two sides of a fault and the geological tectonic background, it is possible to determine whether the fault is a normal fault or a reverse fault. The greater the fault displacement, the more significant the stress redistribution near the fault zone, and the greater the range and extent of its influence on the surrounding area's stress field.

[0074] Furthermore, this embodiment introduces a fault activity rate parameter to quantify fault dynamic activity. Fault activity rate parameter (Unit: mm / day) Defined as the ratio of the cumulative displacement of microseismic events along the fault strike to time within a certain time period:

[0075]

[0076] in, Indicates the first Displacement along the fault slip direction of the microseismic event (unit: mm, calculated from the microseismic positioning deviation). Indicates the statistical period (unit: days); This represents the total number of microseismic events within the statistical time period. The fault activity rate parameter is incorporated into the normal fault fusion model, and the modified formula is:

[0077]

[0078] in, This represents the normal fault weight coefficient (with a value of 1.0, used to distinguish the influence of normal faults from other types of faults). This indicates the fault displacement (unit: m, reflecting the vertical distance between the two sides of the fault). This indicates the dip angle influence factor, based on the fault angle. Division: Take 1; Take 1.3; Take 1.7; The value is 2, which reflects the difference in the disturbance of the stress field by faults with different dip angles; This indicates the vertical distance between the working face and the fault (unit: m; the smaller the distance, the stronger the fault influence). This indicates the fault activity rate (unit: mm / day, reflecting the current dynamic intensity of fault activity).

[0079] Spatial Mathematical Model of Faults: This embodiment uses a spatial geometric coordinate transformation model to describe the spatial location and morphology of faults. Based on the Cartesian coordinate system, it is assumed that the strike vector of the fault is... The tendency vector is The angle of inclination is For any point on the fault It can be obtained through the known coordinates of the fault initiation point. and displacement parameters along strike and dip and To determine its location, the calculation formula is:

[0080] .

[0081] For the quantification of fault length and elevation, spatial distance calculation algorithms in 3D geological modeling software are used. For both ends of the fault... and Fault length The calculation formula is The fault displacement H is calculated based on the vertical distance between corresponding points on both sides of the fault, assuming the coordinates of the corresponding points on both sides of the fault are as follows: and ,but .

[0082] Considering the different mechanical properties of normal and reverse faults, this embodiment further distinguishes them in the model:

[0083] Normal faults: Normal faults are faults in which the hanging wall subsides relatively and the footwall subsides relatively. Their formation is usually related to regional tensile stress. In a spatial geometric coordinate transformation model, when a displacement v occurs along the dip direction, the z-coordinate of the corresponding point on the hanging wall decreases, while the z-coordinate of the corresponding point on the footwall remains unchanged or changes only slightly. Let the initial coordinates of a point on the hanging wall be... Displacement occurs Then, based on the coordinate transformation relationship, the new coordinates The calculation process is as follows:

[0084] exist Axis direction, new coordinate components ;

[0085] exist Axis direction, new coordinate components ;

[0086] exist Axis direction, new coordinate components ,in, The azimuth angle representing the dip angle of the fault.

[0087] The complete three-dimensional coordinate form of the new coordinate system is:

[0088] .

[0089] In this embodiment, a normal fault fusion model is established, and the formula is:

[0090] ,

[0091] in, Fault dip angle (unit: °) refers to the angle between the fault plane and the horizontal plane, which can be obtained directly through geological exploration. The angle between the fault strike and the direction of maximum principal stress (unit: °) is determined through geostress measurement and geological structure analysis. The geological condition influence coefficient ranges from 0.3 to 0.8. It is derived from regression analysis of a large amount of actual data based on factors such as the rock properties and geological history of the mining area. The coefficient of friction of the sliding surface is determined through indoor rock mechanics experiments simulating the fault sliding process, reflecting the roughness and frictional characteristics of the sliding surface. The dip angle of the slip surface (unit: °) refers to the angle between the slip surface and the horizontal plane, obtained through geological surveys and rock structure analysis. The calculation formula for normal faults comprehensively reflects the correction effect of the slip surface characteristics on the surrounding stress field under tensile stress.

[0092] Reverse fault: A reverse fault is a fault in which the hanging wall rises relatively and the footwall falls relatively. Its formation is often related to regional compressive stress. For reverse faults, this embodiment also incorporates the slip surface factor into the model; its formula is:

[0093]

[0094] in, The dip angle of the reverse fault. The angle between the fault strike and the principal stress direction; The vertical stress is (MPa). The maximum horizontal principal stress is (MPa). and The geological condition influence coefficients are denoted by [0.5, 1.2] and [0.2, 0.5], respectively. This formula provides a quantitative description of the changes in the stress field in the reverse fault region caused by the presence of the slip surface.

[0095] (4) Coal seam thickness variation parameters and multiphysics coupling:

[0096] Variations in coal seam thickness are a significant factor influencing the distribution of geostress. This is determined through coal seam thickness data revealed by borehole columnar sections and tunnel excavation. To quantify the degree of change, the calculation formula is as follows:

[0097]

[0098] in, This indicates the thickness of a local coal seam (unit: m, referring to the thickness of a coal seam at a specific location). The average thickness (unit: m, refers to the arithmetic mean of coal seam thickness over the entire study area) indicates the region of coal seam thickness variation. Variations in coal seam thickness lead to uneven gravity distribution in the overlying strata, which in turn affects the geostress state. The greater the rate of variation, the more pronounced the stress anomaly in the corresponding region.

[0099] Different stress adjustment coefficients k correspond to different rates of change in coal seam thickness.

[0100] when At that time, the coal seam thickness changed relatively gently, and the stress distribution was relatively uniform. ;

[0101] when At that time, changes in the thickness of the coal seam can cause local adjustments in stress. ;

[0102] when At that time, the coal seam thickness varied drastically, and the stress distribution differed significantly. .

[0103] The vertical and horizontal stresses are corrected using stress adjustment factors, and the correction formulas are as follows:

[0104]

[0105]

[0106] in, This represents the original vertical stress (unit: MPa, vertical stress without thickness variation correction). This represents the original horizontal stress (unit: MPa, horizontal stress without thickness variation correction). This represents the corrected vertical stress (unit: MPa). This represents the corrected horizontal stress (unit: MPa).

[0107] For high-gas mines, changes in coal seam thickness can also couple with factors such as gas pressure. Therefore, this embodiment increases the fold-gas pressure coupling coefficient. (Unit: 1 / m), establish a correlation model between fold curvature and gas pressure:

[0108]

[0109] in, The permeability coefficient of coal (unit: m² / (MPa·s) reflects the ability of coal to allow gas flow; the higher the value, the easier it is for gas to flow and diffuse). This indicates the coal seam gas pressure (unit: MPa, referring to the pressure of gas in the coal seam). This represents the radius of curvature of the fold (unit: m, reflecting the degree of fold curvature; the smaller the value, the more severe the fold deformation). As a correction term for the initial geostress field, the geostress is corrected, and the specific correction formula is as follows:

[0110]

[0111] in, This represents the initial geostress (unit: MPa, geostress value before coupling correction). This represents the corrected in-situ stress (unit: MPa, the stress value after comprehensively considering mechanical and gas factors). This correction comprehensively considers the combined effects of mechanical factors and gas pressure, solving the problem of traditional models only considering mechanical factors and neglecting the influence of gas pressure. When the value is large, it indicates that the folding deformation and gas pressure have a significant impact on the initial geostress field, requiring a substantial correction.

[0112] Step 1.2: Obtain measured parameters of downhole in-situ stress, including: measured data of triaxial principal stress, stress gradient, and six stress components obtained based on hydraulic fracturing or stress relief methods.

[0113] In this embodiment, the step specifically includes:

[0114] (1) Conventional stress measured parameters: The measured values ​​of the three principal stresses, including the vertical principal stress, were obtained by using the hydraulic fracturing method and the stress relief method. Horizontal principal stress and Simultaneously, acquire stress gradient data, such as vertical stress gradient. Horizontal stress gradient and And geostress direction data (principal stress azimuth and dip angle).

[0115] (2) Measurement of six-dimensional stress components: Based on the principle of strain sensing, spherical stress sensors are distributed in the surrounding rock of coal mine roadways or in boreholes. The sensors can collect the three-dimensional stress state in space in real time and directly obtain six stress components: three normal stress components. , , and three shear stress components , , Simultaneously, it records time-series data of stress changes, with a sampling frequency of up to 10Hz and a measurement accuracy of ±0.05MPa, capable of capturing instantaneous stress fluctuations. The complementary verification between the six stress components acquired by the spherical stress sensor and the measured triaxial principal stress data provides more accurate measured data for establishing the initial geostress field.

[0116] (3) Principles for the arrangement of stress measurement points: When obtaining the measured values ​​of the three principal stresses, the number and arrangement of measurement points should follow the following principles:

[0117] ①Basic quantity requirements: For a typical deep coal mine, at least 8 triaxial principal stress measurement points should be arranged in the initial geostress field modeling stage, evenly distributed in different areas of the mine, covering the main mining level, different geological structure areas (such as near faults, fold areas) and areas with different coal seam occurrence conditions (such as coal seam thickness variation zones).

[0118] ② Densification in areas with complex geological structures: In areas with complex geological structures (such as fault zones and fold cores), the spacing between measuring points should be increased to 50-100m in order to capture the characteristics of tectonic stress concentration.

[0119] ③ Key monitoring of coal seam change areas: In areas where the coal seam thickness change rate is greater than 20% or in areas of abrupt coal quality change (such as the soft-hard coal transition zone), special monitoring points should be set up, with at least one monitoring point set up in each change area, in order to study the impact of changes in coal seam occurrence conditions on stress distribution.

[0120] ④ Dynamic adjustment mechanism for the number of measuring points: If the initial set of measuring point data has large dispersion (stress value deviation of measuring points in the same area exceeds 15%), or the model verification error exceeds the threshold, 1-2 additional measuring points should be added to that area to supplement the data. When the improvement in model accuracy by the added measuring points is less than 5%, the number of measuring points can be considered sufficient.

[0121] Step 1.3: Acquire data on the direction of geostress, anomalies, and correlation characteristics of tectonic stress energy.

[0122] Specifically, comprehensive data on the direction of in-situ stress, stress anomalies, and the correlation characteristics of tectonic stress energy are collected, including: analyzing the morphology and combination of faults, folds, and fracture zones based on 3D seismic and trough wave exploration; qualitatively determining the dominant direction of local tectonic compression and shear stress through tectonic deformation characteristics, fault strikes, and fold hinge attitudes; identifying areas of high in-situ stress concentration based on 3D seismic wave velocity, wave impedance, and reflection energy parameters; and obtaining the source rupture direction, event distribution, and cumulative energy parameters through microseismic monitoring. In this embodiment, this step specifically includes:

[0123] (1) Based on the results of 3D seismic exploration and channel wave exploration, the structural morphology and spatial combination of faults, folds, and fracture zones within the well field are analyzed. By analyzing structural deformation characteristics (such as the degree of fold tightness and the gradient of stratigraphic dip), the dominant strike of faults, and the attitude of fold hinges, the dominant directions of tectonic compression and shear stresses in the local area are qualitatively determined. This step provides a geological basis for setting stress boundary conditions in subsequent numerical simulations.

[0124] (2) Using the seismic wave data acquired by three-dimensional seismic exploration, the P-wave velocity of the coal and rock mass is obtained by inverting the first arrival travel time. and transverse wave velocity Regions of velocity anomalies (such as sudden decreases or increases in velocity) suggest changes in the mechanical properties of the rock mass, potentially indicating stress concentration. Simultaneously, migration imaging is performed on the seismic reflected waves to extract attributes such as amplitude, phase, and frequency. In stress concentration zones, rock mass structural failure leads to weakened reflected wave energy and dislocation or distortion of the phase axis; these characteristics are used to determine the boundaries and extent of the stress concentration zones.

[0125] (3) By deploying a micro-seismic sensor array in the coal mine roadway and working face, the time difference of the arrival of micro-seismic waves at each sensor is collected. Combined with wave velocity in coal and rock mass Using the hyperbolic positioning method or the minimum travel time inversion algorithm, the three-dimensional coordinates of microseismic events are calculated. The formula is: ,in To determine the propagation distance of microseismic waves, through multiple sets and Solving the simultaneous equations to find the location coordinates allows for precise localization of microseismic events, directly reflecting the location distribution of stress concentration areas. Simultaneously, analyzing the spectral characteristics of the microseismic waveform allows for the calculation of parameters such as source energy E and magnitude M. The formula for calculating source energy is as follows: Where A is the waveform amplitude, ω is the angular frequency, r is the distance from the source to the sensor, and Q is the medium quality factor in the formation. Areas with concentrated high-energy, high-magnitude microseismic events indicate a high degree of stress concentration in the region and can be used as one of the criteria for determining stress concentration zones in the initial geostress field model.

[0126] Step 1.4: Using the mine geological parameters, coal and rock mechanical parameters, and geostress concentration zone parameters as inputs, perform simulation calculations using finite element or discrete element numerical models to generate initial geostress field simulation results.

[0127] Specifically, refined geological parameters, coal and rock mechanics test parameters, geostress direction characteristic parameters, and geophysical stress anomaly parameters are uniformly input into a finite element or discrete element numerical model. Combined with in-situ measured geostress data, the model boundary stress magnitude and direction are constrained to reconstruct the stratigraphic compression and shear mechanics mechanisms. A full-domain, multi-scale geostress numerical simulation is conducted, and an elastic strain energy evaluation model is simultaneously constructed. By combining seismic wave velocity, mechanical parameters, and monitoring energy data, the stress accumulation intensity, accumulated energy level, and spatial distribution of the surrounding rock are quantified. Finally, a full-domain initial geostress field simulation result containing stress values, stress directions, and strain energy partitions is obtained. In this embodiment, this step specifically includes:

[0128] (1) Multi-source data integration and model construction: All parameters obtained in steps 1.1 to 1.3 are deeply integrated and verified, and input into the same numerical simulation system. The numerical simulation method can be the finite element method, the discrete element method, or their coupling method. In the modeling process, a numerical grid is constructed based on the geological structure model (including the geometric and mechanical properties of faults and folds) established in step 1.1, and the coal and rock mechanical parameters are assigned to the corresponding stratigraphic units.

[0129] (2) Boundary condition setting and mechanism restoration: Based on the in-situ geostress measurement data, the magnitude and direction of the boundary stress of the model are set. The key point is that it is not simply to apply a uniform stress boundary, but to restore the tectonic compression and shear stress dominance direction based on the qualitative analysis in step 1.3, so that the simulated stress field not only conforms to the measured data, but also is consistent with the mechanical genesis of the geological structure.

[0130] (3) Simultaneous construction of elastic strain energy evaluation model: While carrying out numerical simulation of geostress, an elastic strain energy evaluation model is constructed simultaneously. This model combines seismic wave velocity, mechanical parameters, and monitoring energy data to quantify the stress accumulation intensity, energy storage level, and spatial distribution of the surrounding rock.

[0131] (4) Simulation Results Output: After completing the numerical calculation, the simulation results of the global initial geostress field, including the following three types of information, are output: ① Stress values ​​(stress tensor, principal stress magnitude and direction at each node); ② Stress direction (principal stress vector distribution); ③ Strain energy zoning (energy level zoning map based on the elastic strain energy evaluation model). Through numerical simulation technology, the distribution characteristics and evolution of the stress field under complex geological conditions and mining processes can be presented intuitively in a visual form, providing solid data support and theoretical basis for deep coal mine mining design and disaster prevention.

[0132] Step 1.5: Compare and verify the simulation results of the initial geostress field across the entire region with the measured geostress data. Iterate and correct the model parameters until the simulation result error is lower than the preset threshold, and generate the initial geostress field model.

[0133] Specifically, by comparing and verifying the numerical simulation results with measured data, the differences between the simulation results and the measured data are analyzed to identify the causes of errors. If the error is caused by unreasonable model parameter settings or an imperfect model structure, the mathematical model is corrected and optimized by adjusting model parameters or improving the model structure until the error between the simulation results and the measured data is within an acceptable range. Through continuous comparison, verification, and correction, the accuracy and reliability of the mathematical model are improved. In this embodiment, this step specifically includes:

[0134] (1) Error evaluation index: The numerical simulation results are compared and verified with the measured data. The accuracy of the model is evaluated by calculating the root mean square error (RMSE) and mean absolute error (MAE).

[0135]

[0136]

[0137] in, This represents the geostress value obtained from numerical simulation. This represents the measured ground stress value, and n represents the number of comparison data points.

[0138] (2) The specific settings for revising the trigger threshold are as follows:

[0139] ① Stress magnitude error threshold: For triaxial principal stresses, when RMSE is greater than 1.5 MPa or MAE is greater than 1.0 MPa, the model error is considered to exceed the acceptable range, and the numerical model parameters need to be adjusted. This is because in the stress environment of deep coal mines, a stress error greater than 1.5 MPa may lead to misjudgment of high-stress danger areas, affecting the accuracy of disaster early warning; while a MAE exceeding 1.0 MPa indicates that the overall deviation of the model is large and cannot meet the needs of engineering applications.

[0140] ② Stress direction error threshold: When the root mean square error (RMSD) of the principal stress direction deviation angle exceeds 3° or the mean absolute error (MAD) exceeds 3°, the model parameters need to be readjusted. The stress direction has a significant impact on the rock mass failure mode and disaster occurrence mechanism. A large direction error may lead to incorrect judgment of potential stress concentration paths and failure surfaces.

[0141] ③ Comprehensive accuracy threshold: When the simulated stress value in the stress concentration zone deviates from the measured value by more than 5%, or the location deviation of the stress concentration zone exceeds 5% of the actual range, regardless of whether indicators such as RMSE and MAE meet the standards, the model parameters must be adjusted or the structure optimized. This is because the stress concentration zone is a key area for dynamic disasters, and higher requirements are placed on its simulation accuracy.

[0142] (3) Iterative correction: If the error exceeds the set threshold, the numerical model parameters (such as grid density and constitutive model parameters) are adjusted and the simulation is repeated until the simulation results are in high agreement with the measured data, and finally a high-precision initial geostress field distribution model is generated.

[0143] In one embodiment, after completing the initial geostress field modeling in step 1, this embodiment further provides a method for evaluating dynamic disaster sensitivity indicators in step 2. The sensitivity indicator evaluation system in this embodiment is guided by the "Detailed Rules for the Prevention and Control of Coal Mine Rockbursts" and relevant requirements for gas control in Henan Province. Based on the "Detailed Rules for Prevention and Control of Rockbursts," stress and coal-rock physical and mechanical parameters are selected to construct a scientific disaster sensitivity evaluation system, providing key evidence for early warning of dynamic disasters. A sensitivity indicator system is established from four dimensions: stress state, geological structure, coal-rock characteristics, and gas-related indicators, and corresponding evaluation methods are formulated, specifically including:

[0144] (1) Stress state sensitive indicators include the maximum principal stress concentration factor, stress gradient and stress-gas coupling hazard index, which are used to quantify the concentration degree, variation gradient and multi-physics coupling effect of the surrounding rock stress field under mining disturbance.

[0145] Among them, the maximum principal stress concentration factor ( Based on the established initial and dynamic stress field data, the ratio of the maximum principal stress to the original rock stress in the area surrounding the mining face or roadway is calculated using the following formula:

[0146]

[0147] in, The maximum principal stress is expressed in megapascals (MPa). This refers to the original rock stress, measured in megapascals (MPa). When the coal seam is buried at a depth exceeding 400 meters, and within 100 meters above the coal seam exists a single, hard rock layer exceeding 10 meters in thickness and with a uniaxial compressive strength greater than 60 MPa, and simultaneously the gas pressure is greater than 0.6 MPa or the gas content is greater than 6 cubic meters per ton, if... Special attention needs to be paid to the risk of rockbursts (compared to normal conditions, the risk warning threshold should be lowered and the monitoring level increased under these conditions). Based on measured data and historical disaster cases, [the following will be implemented]. Divided into four risk levels: Low risk Medium risk. High risk, This is extremely high risk.

[0148] Stress gradient ( The formula reflects the rate of change of stress with spatial location and is as follows:

[0149]

[0150] in, The stress difference between adjacent monitoring points is expressed in megapascals (MPa). This represents the distance between two points, expressed in meters (m). When the coal seam is at the aforementioned burial depth, rock strata, and gas conditions, if... We need to immediately strengthen the monitoring of dynamic disasters, lower the stress gradient early warning threshold and increase the monitoring frequency compared to ordinary coal seam conditions.

[0151] The stress-gas coupling hazard index C is used to quantify the synergistic effect of stress and gas. The calculation formula is as follows:

[0152]

[0153] in, This indicates the maximum principal stress (unit: MPa, referring to the maximum principal stress value acting on the rock mass). It represents the compressive strength of coal (unit: MPa, the ultimate ability of coal to resist crushing). Indicates the measured gas pressure (unit: MPa); This indicates the critical pressure for coal and gas outbursts (unit: MPa, taken as 0.74 MPa, the critical gas pressure value for coal and gas outbursts). When When the risk level is high, the existing indicators fail to quantify the synergistic effect of stress and gas.

[0154] (2) Geological structure sensitive indicators include fault influence coefficient and fold curvature index, which are used to evaluate the degree of influence of geological structures such as faults and folds on dynamic disasters.

[0155] Among them, the fault influence coefficient A fault influence coefficient model is constructed by comprehensively considering fault type, elevation difference, distance from the mining face, and fault dip angle. Weighting coefficients are assigned to normal and reverse faults respectively. and Introducing fault dip angle influence factor (Unit: degrees, °) The fault dip angle The judgment is divided into four intervals: when hour, The value is 1; when hour, The value is 1.3; when hour, The value is 1.7; when hour, The value is set to 2. The formula for calculating the fault influence coefficient is:

[0156]

[0157] Where h is the fault elevation, in meters (m); This is the vertical distance between the working face and the fault, expressed in meters (m). When When a fault zone is identified as a high-impact fault zone, strict dynamic disaster prevention measures must be implemented, such as production shutdowns for rectification, increased support strength, and shortened monitoring cycles. By refining the dip angle range, the differences in the impact of faults with different dip angles on dynamic disasters can be reflected more accurately. For example, high-dipping reverse faults, compared to low-dipping faults, have a more significant impact on the surrounding stress field, giving it a higher [impact]. This is to reflect its greater influence.

[0158] Wrinkle curvature index Based on 3D seismic exploration data, the least squares method was used to fit the rock strata surfaces of synclines and anticlines, and the rock strata curvature was calculated through the surface differential equations. For two-dimensional curves... The specific formula for curvature is:

[0159]

[0160] In three-dimensional space, by constructing the equation of the rock strata surface Calculate principal curvature using partial derivatives , The larger value is taken as the fold curvature of the region. The unit is meters. ( Based on the magnitude of the fold curvature, The risk level determination criteria have been adjusted as follows:

[0161] when hour, This indicates that the folds are gentle and the formation stress is relatively uniform.

[0162] when hour, It has a medium curvature; hour, The folds have a high curvature.

[0163] when hour, The curvature is significant;

[0164] when hour, It has extremely high fold curvature.

[0165] Simultaneously, based on microseismic monitoring data, if the frequency of microseismic events in the folded region suddenly increases by more than 20% and the average energy release increases by 15% within a certain time period, then the current... The risk level has been raised by one level. By inverting fold morphology using 3D seismic data and combining this with numerical simulation analysis of stress distribution characteristics under different curvatures, a system was established. The correlation between the probabilities of dynamic disasters provides a reliable basis for disaster early warning.

[0166] (3) Sensitive indicators of coal and rock properties include media impact tendency index and rock mass brittleness index, which are used to evaluate the mechanical properties of coal and rock mass itself and its tendency to dynamic disasters.

[0167] Among them, the coal seam impact tendency index A comprehensive index method was adopted, combining the dynamic failure time (DT) of coal (in milliseconds, ms) and the elasticity index. Impact Energy Index The calculation formula is as follows:

[0168]

[0169] in, , , These are the weighting coefficients; The standard dynamic failure time is expressed in milliseconds (ms). It needs to be based on... The value strictly classifies the coal seam's shock tendency into four levels: none, weak, moderate, and strong. For coal seams under special conditions, if... (For cases with a moderate tendency to rockburst), a rockburst monitoring and mitigation plan should be initiated immediately.

[0170] Rock mass brittleness index Uniaxial compressive strength of rock (Unit: megapascals, MPa) and tensile strength The ratio (unit: megapascal, MPa) is characterized by the following formula:

[0171] .

[0172] When the coal seam meets the relevant conditions, if We must be highly vigilant about the occurrence of dynamic disasters such as rock bursts. In this case, compared with ordinary coal seam conditions, we should lower the rock mass brittleness index warning threshold and take rock burst prevention measures in advance.

[0173] (4) Gas-related sensitive indicators include gas pressure, gas content, gas analysis index, initial velocity of borehole gas outburst, amount of drill cuttings and rock mass damage evolution coefficient, which are used to evaluate the correlation between the occurrence state of coal seam gas and its dynamic disasters.

[0174] Among them, gas pressure A direct measurement method was employed, with pressure-measuring boreholes installed in the coal seam. Gas pressure was accurately measured using a sealing device and pressure sensors. When the gas pressure... When the pressure is high, it indicates that the coal seam gas is under high pressure, the risk of dynamic disasters is significantly increased, and mining operations must be stopped immediately. Enhanced gas extraction measures should be implemented, and the frequency of gas pressure monitoring should be increased to ensure that the pressure drops below the safe threshold.

[0175] Gas content The gas content of the coal seam is calculated by combining geological survey data with on-site sampling analysis. When the gas content... At that time, the coal seam gas was abundant, which could easily cause dynamic disasters such as gas outbursts. It was necessary to carry out comprehensive gas control in the area, such as using surface gas decompression drilling, cross-layer drilling and hydraulic cavity creation, to enhance the gas extraction effect and reduce the gas content.

[0176] Gas analysis index ( or According to the "Detailed Rules for Preventing Gas Outbursts," gas desorption parameters are measured in boreholes using specialized instruments at the coal roadway excavation face. A commonly used parameter is the gas desorption parameter from drill cuttings. (The unit is the square root of milliliters per gram per minute) )or (Unit: Pascal). When the corresponding conditions are met, if or The area was determined to have a prominent risk tendency, and gas control measures should be strengthened immediately, such as increasing the intensity of extraction and increasing the desorption time.

[0177] Initial velocity of gas ejection from borehole According to the MT / T 639-2019 standard, the maximum gas flow rate within 2 minutes is measured in a borehole of a specified length (1.0 m) in a coal seam, expressed in liters per minute (L / min). When the corresponding conditions are met, if... This area should be considered a high-risk area, and measures such as supplementary gas extraction and enhanced ventilation should be taken immediately to reduce the risk of gas outbursts.

[0178] Drill cuttings In coal roadway excavation faces, all drill cuttings are collected after every meter of drilling, and the weight is determined using either a gravimetric method (weighing with a spring scale) or a volumetric method (measuring the volume of drill cuttings with measuring instruments), with the unit being kilograms per meter (kg / m). Under specific conditions, if... If the working face is determined to have a gas outburst hazard, tunneling operations must be stopped and special gas outburst prevention measures implemented, such as pressure relief blasting and reinforcement of support.

[0179] To incorporate damage accumulation over time into the evaluation, this embodiment constructs a rock mass damage evolution coefficient D, calculated using the following formula:

[0180]

[0181] in, This indicates the number of microseismic events within 30 days (unit: events). This indicates the historical average number of micro-seismic events (unit: times, referring to the average number of micro-seismic events during the same period in the past year). This represents the cumulative energy of microseismic events over 30 days (unit: J). This represents the historical average cumulative microseismic energy (unit: J, referring to the average microseismic energy over the same period in the past year). When D > 2.5, it indicates accelerated rock mass damage and a significantly increased risk of dynamic disasters. This indicator incorporates the cumulative damage over time into the evaluation, overcoming the shortcomings of existing indicators that only reflect instantaneous conditions.

[0182] (5) Comprehensive evaluation model

[0183] After normalizing the aforementioned sensitive indicators, the weights of each indicator were determined using the Analytic Hierarchy Process (AHP) combined with expert scoring. A comprehensive evaluation function was then constructed. ,in Let be the weight of the i-th indicator; This is the normalized index value.

[0184] In this embodiment, the step of determining the indicator weights includes:

[0185] An expert evaluation panel of 5-10 senior experts in coal mine geology, mining engineering, and gas control will be formed. Quantitative scoring indicators will be developed, evaluating indicators based on three dimensions: the degree of impact on dynamic disasters, the reliability of monitoring data, and the operability of the indicators in actual engineering projects. Each dimension will have a maximum score of 10 points.

[0186] For four dimensions—stress state, geological structure, coal and rock properties, and gas-related indicators—and the specific indicators under each dimension, experts scored the relative importance of the indicators using a 1-9 scale (1 indicates that two indicators are equally important, and 9 indicates that one indicator is extremely important than the other), thus constructing a judgment matrix. ,in This indicates the relative importance of indicator i and indicator j.

[0187] Calculate the weight vector: Calculate the eigenvectors of the judgment matrix A using the root method or the sum-product method, and then normalize the eigenvectors to obtain the weight vectors of each indicator. .

[0188] At the same time, a consistency check is performed, and a consistency index is calculated. The random consistency index RI (obtained by looking up a table based on the matrix order) is the consistency ratio. If the judgment matrix is ​​considered to have satisfactory consistency, the weight vector is considered valid; otherwise, the judgment matrix needs to be readjusted through expert scoring. This represents the largest eigenvalue of the judgment matrix A.

[0189] Furthermore, in order to adapt to the dynamic changes in disaster-affecting factors at different mining stages underground, this embodiment adopts a data-driven dynamic weight generation algorithm based on the base entropy of the above steps.

[0190] Specifically, based on historical disaster case data (records of mine dynamic disasters and corresponding monitoring data over the past 10 years), an LSTM deep learning model is constructed. Using "indicator value - probability of disaster occurrence" as training samples, the model automatically learns the indicator weights for different mining stages. For example, within 50m of the tunneling face, the model can automatically increase the weight of "forward stress gradient" from 0.2 to 0.35; after entering the goaf's influence range, the weight of "roof and floor stress difference" is increased from 0.15 to 0.25, replacing the manually set dynamic weight function.

[0191] To ensure the scientific validity and reliability of the weights, a weight credibility verification mechanism is incorporated. 1000 sets of random weights are generated through Monte Carlo simulation, and the degree of agreement between the evaluation results under each weight set and the actual disaster is calculated. The formula for calculating the degree of agreement is as follows:

[0192]

[0193] Only weighted ranges with a fit greater than 85% are retained. For example, the stress concentration factor weight is limited to 0.18-0.22 after verification to ensure the scientific validity of the weighting.

[0194] Finally, based on the R-value and relevant specifications, the risk level of dynamic disasters can be classified, and corresponding early warning thresholds and prevention and control strategies can be formulated. In this embodiment:

[0195] when At that time, it was considered low risk;

[0196] At that time, it was classified as medium risk;

[0197] At that time, it was considered high-risk;

[0198] At that time, it was considered extremely high risk.

[0199] For coal seams that meet the criteria for identifying shock tendency as stipulated in the detailed rules, or coal seams whose gas-related indicators reach the threshold, when R reaches the corresponding threshold, a detailed identification procedure and enhanced prevention and control measures shall be immediately initiated to ensure safe production in coal mines.

[0200] During coal mining, the stress field dynamically changes as mining activities proceed. In one embodiment, to achieve effective early warning of dynamic disasters, stress detectors and microseismic detectors are deployed throughout the mine, around the mining area, coal face, and tunneling face, to acquire stress data in real time and correct the initial stress field. The implementation details of step 3 specifically include:

[0201] In one embodiment, to achieve dynamic perception of the stress field throughout the mine, equipment is deployed at three levels: the tunneling face, the coal mining face, and the mining area. Spherical stress detectors and microseismic detectors are used to dynamically acquire data such as stress magnitude, direction, and microseismic signals in real time.

[0202] At the tunneling face, spherical stress detectors are deployed in a triangular array on both sides and the roof of the tunnel, 5-10 meters behind the tunnel face. This area is sensitive to stress concentration and changes; this layout allows for comprehensive capture of the magnitude and direction changes in stress in the surrounding rock in front of and around the tunneling face. Microseismic detectors are installed in stable rock masses 10-15 meters from the tunnel face, with one detector every 15-20 meters along the tunnel axis. These detectors are used to monitor microseismic signals generated by rock mass fracturing in real time, in order to determine the degree of stress concentration and potential fracturing risk.

[0203] In the coal mining face, spherical stress detectors are installed at the connection between the top beam and base of the hydraulic support, and inside the coal wall at both ends of the working face. The former can sense the stress changes of the roof and floor plates of the working face in real time, while the latter can monitor the stress state of the end areas. Microseismic detectors are arranged deep in the coal wall of the return airway and transport roadway of the working face, with one installed every 30-50 meters, forming a monitoring network covering the entire working face.

[0204] At the mining area level, spherical stress detectors are deployed at the main roadway intersections, boundary coal pillars, and areas with complex geological structures. At roadway intersections, detectors are placed at intervals of 3-5 meters around the intersection on the surrounding roadway walls to monitor stress superposition changes at the intersection. At boundary coal pillars, a detector is installed every 20-30 meters along the pillar's strike to monitor the pillar's bearing stress. In areas with complex geological structures, such as near faults and folds, detectors are deployed more densely, with intervals reduced to 10-15 meters, to track stress anomalies in the structural areas in real time. Microseismic detectors are evenly distributed in the roof strata of the mining area, with a monitoring point set every 50-80 meters, forming a three-dimensional monitoring network.

[0205] Stress detectors can accurately measure the magnitude and direction of stress at different locations in deep coal mines. By strategically deploying n spherical stress detectors within the mine, a comprehensive stress monitoring network can be constructed. The stress tensor measured by the stress detector at time t (s, seconds) is: In Cartesian coordinates, it can be represented as:

[0206]

[0207] Among them, each component of the stress tensor The unit is Pa (Pascal). Considering measurement error, the actual measured value... Compared with the true value Existence Relationship: ,in To measure the error matrix, the units are the same as those of the stress tensor. Error correction is performed by taking the average of multiple measurements.

[0208] The waveform function of the microseismic signal is By analyzing the spectrum of the microseismic signal, the frequency components of the signal can be obtained. (Hz, Hertz) and its amplitude Based on microseismic location algorithms, such as the hyperbolic location method, the locations of microseismic events are established. Mathematical model:

[0209]

[0210] Wherein, min represents the value obtained by adjusting the location of the microseismic events. The value of makes the expression on the left side of the equation reach its minimum value, at which point the corresponding value is... This represents the optimal estimated location of the microseismic event. This principle can be analogized to finding a point on a map that minimizes the sum of the differences between that point and multiple known locations calculated according to certain rules (here, wave speed and propagation time). Specifically, the expression on the left-hand side of the equation calculates the location of the microseismic event based on the assumed location. The calculated theoretical time for microwaves to propagate to each monitoring station differs from the actual measured time for microwaves to arrive at each station. The sum of squared differences between the two. When this sum of squared differences is minimized, it means that the propagation time calculated from the assumed location is closest to the actual measured time. This is the location where a microseismic event is most likely to actually occur.

[0211] In one embodiment, the real-time acquired data is compared and analyzed with the initial geostress field established before mining. Let the stress vector of the initial static geostress field be... The stress vector acquired in real time is .in, , , These represent the stress values ​​of the initial static geostress field in the x, y, and z directions, respectively, in megapascals (MPa). , , These represent the stress values ​​acquired in real time along the x, y, and z directions, respectively, with the unit also being megapascals (MPa). Construct the difference function:

[0212]

[0213] in, The weights for stress data in each direction are given, and their values ​​range from [value range missing]. It is dynamically adjusted according to different mining areas (such as tunneling faces, coal mining faces, and specific locations in mining areas) and geological conditions (such as near faults, folds, and other structures) to highlight the importance of stress data in different directions in the overall difference calculation.

[0214] It is understandable that the difference function serves as a crucial bridge connecting "real-time data comparison" and "model correction." Its core function is to quantitatively compare the initial static stress vector with the real-time acquired stress vector by introducing directional weights. ( (Adaptable to different mining areas / geological conditions), accurately calculating the degree of stress difference in each spatial direction. And locate the area of ​​abnormal stress.

[0215] In a preferred embodiment, geological structural features are also incorporated to aid in determining the direction of geostress. Since the morphology, orientation, and occurrence of geological structures such as faults, folds, and joints are direct results of long-term regional geostress, there is a fixed correspondence between structural morphology and geostress direction. Therefore, based on the distribution of geological structures, the direction of geostress can be determined in two ways: first, by artificially calibrating the geostress direction according to the structural occurrence and stress characteristics; and second, by using structural parameters to invert and solve for the geostress direction.

[0216] Based on the comparison results, an adaptive data correction algorithm is used to dynamically correct the initial stress field. A stress field parameter update model is constructed, and the corrected stress vector is assumed to be... ,but ,in Let this be the stress variation vector, as described above. To standardize the physical parameters, the units for the components in the x, y, and z directions are all megapascals (MPa). This is a correction factor, and its value range is... It is dynamically determined based on the stress variation amplitude, direction, and geological structure evolution, and is used to control the magnitude of stress vector adjustment during each correction.

[0217] For areas near faults, folds, and other special geological structures, the weighting is increased. and adjustment correction factor The weighting of monitoring data in this area will be increased to highlight stress concentration and anomalies. Simultaneously, the data weighting will be dynamically optimized based on different mining stages (tunneling, longwall mining, etc.) and the overall stress variation patterns of the mining area. and correction factor A dynamic stress field model tailored to the current mining conditions is constructed. During the correction process, the direction of in-situ stress is adjusted based on calibration or inversion results, and the stress values ​​are iteratively corrected through model inversion and estimation. Simultaneously, a stress field parameter update model is constructed to verify the rationality of the corrected stress vector. After passing the verification, iterative updates are performed to establish a new benchmark stress field and stress vector, achieving dynamic adaptive correction of the stress field. This provides precise ground stress foundation support for the evaluation of multi-dimensional sensitive indicators, deep fusion of multi-source data models, and four-level early warning and prevention, forming a closed-loop process of benchmark establishment—dynamic correction—fusion analysis—early warning and prevention.

[0218] In one embodiment, the large-scale model constructed in step 4 achieves deep fusion of multi-source data and dynamic disaster analysis and early warning through a mathematical model. The core of the large-scale model in this embodiment lies in constructing a graph neural network fusion model with dual-layer weighted collaborative fusion. Dynamic stress field model data, sensitive index evaluation system data, and downhole real-time monitoring data are used as model inputs. The model completes the fusion of measurement point space and geological features based on the graph neural network message passing mechanism, and completes secondary feature weighting optimization by combining dynamic weights of working conditions. The final fused features are input into the early warning level determination model, and the dynamic disaster early warning result is output through fuzzy membership degree synthesis operation.

[0219] In one embodiment, the collected multi-source data set is: ,in Representing different types of data: This represents geostress field data (unit: MPa, megapascal), including parameters such as maximum principal stress and minimum principal stress; This represents microseismic monitoring data (unit: times / minute, used to characterize the frequency of microseismic events). Represents geological structural data (such as fault drop ratio, normalized value of fold curvature); Data representing the mechanical properties of coal (rock) (unit: depending on the specific parameter, such as elastic modulus in GPa).

[0220] To rationally utilize data from different monitoring methods during the fusion process, data fusion weights need to be constructed to measure the credibility and importance of data from each source. The determination of data fusion weights employs a combination of the Analytic Hierarchy Process (AHP) and the entropy weight method. Subjective data weights are determined by domain experts based on their experience in deep coal mining and the perceived importance of the data, using a judgment matrix constructed using the AHP method. Objective data weights are calculated using the entropy weight method based on the inherent variability of the data. The final data fusion weights are calculated using the following formula:

[0221]

[0222] in, For corresponding data The weights, and satisfying (Dimensionless, weights are proportional values). Weight The determination of the value is achieved using a combination of the Analytic Hierarchy Process (AHP) and the entropy weight method: This is a balance coefficient (dimensionless, ranging from [0,1]), used to balance the influence of subjective experience and objective data. When the weights are determined entirely by the objective characteristics of the data; when At this time, the weights rely entirely on expert subjective judgment. Subjective weights are determined by domain experts based on their experience in deep coal mining and the importance of the data. An AHP (Analytic Hierarchy Process) method is used to construct a judgment matrix to determine the initial subjective weights. Objective weights are calculated using the entropy weight method based on the degree of variation in the data itself. It should be noted that the aforementioned data fusion weights are comprehensive basic weights applicable to a general data fusion scenario across the entire mine. Based on this, this embodiment further designs a dynamic weight adjustment mechanism for the core risk data of two mining scenarios: tunneling faces and coal mining faces, to perform secondary scenario-based optimization of the comprehensive weights.

[0223] For the tunneling face, the weighting of the stress variation data ahead. The dynamic weight adjustment function will be dynamically adjusted based on the working face advancement distance L (unit: meters, m). :

[0224]

[0225] in, As the initial weights, It is a monotonically increasing function of the propulsion distance L, for example , , This is an adjustment factor (dimensionless). This represents the total advancing length of the working face (unit: meters, m). This function gradually increases the weight of the forward stress data as the tunneling distance increases, highlighting its importance in the tunneling process.

[0226] For coal mining faces, data considering periodic pressure and changes in key roof strata are used. Construct a weight adjustment function. Let the periodic pressing step distance be S (unit: meters, m), and the ratio of the actual advancing distance to the periodic pressing step distance be... (Dimensionless), the thickness variation of the key layer of the top plate is: (Unit: meters, m), the initial thickness of the key layer of the top slab is... (Unit: meters, m), then the weight The adjustment function is:

[0227]

[0228] in, For data The initial weights (dimensionless) are determined by a combination of the analytic hierarchy process (AHP) and the entropy weight method. This is a weight adjustment function whose output value depends on r and It changes with the changes in the two parameters, and the specific function form is as follows:

[0229]

[0230] in, , , The adjustment coefficients (dimensionless) were determined through extensive experimental data and model training optimization, and are used to adjust the degree of influence of different parameters on the weights; The periodic pressure characteristic ratio (dimensionless) represents the r value when the periodic pressure is most significant. In the function, The term is used to quantify the effect of the periodic pressure: when r approaches That is, when the working face approaches the periodic pressing stage, the value of this function increases rapidly, making... This significantly increases the weight of data related to periodic pressure in the fusion process; The term is used to reflect the impact of changes in the thickness of the critical layer of the roof on the weights; the amount of change in the thickness of the critical layer of the roof. The larger the value, the higher the corresponding data weight. The higher the weight, the more dynamic the data fusion weights of the coal mining face are optimized. It should be noted that the dynamic weight adjustments for the tunneling and coal mining faces described above are specific adjustment rules for core risk data, not simply examples, but essential dynamic optimization steps in the entire weight calculation process. The dynamic weight adjustments for both scenarios only apply to core risk data; the weights for other non-core data still use the comprehensive basic weights, thus achieving precise matching of data fusion weights with the risk characteristics of different mining scenarios.

[0231] It is understandable that the weights of tunneling faces and coal mining faces are dynamic and change with mining operations. These weights represent a secondary optimization of the core risk data for both mining scenarios based on the comprehensive weights. The comprehensive weights, as a general foundation for multi-source data fusion, are calculated by harmonizing subjective weights determined by the analytic hierarchy process (AHP) and objective weights determined by the entropy weight method using a balance coefficient. This comprehensive weight is applicable to a universal data fusion scenario across the entire mine. For tunneling faces, the comprehensive weights of the stress change data ahead are used as a benchmark. A monotonically increasing adjustment function, positively correlated with the face's advance distance, is used to dynamically increase the weights, thus highlighting the core risk of stress concentration ahead during tunneling. For coal mining faces, the comprehensive weights of the periodic pressure and roof key layer change data are used as a benchmark. An adjustment function, relating the advance distance to the periodic pressure step distance and the roof key layer thickness change rate, is used to dynamically change the weights. This strengthens the influence of roof stress-related data during the periodic pressure stage. Furthermore, the dynamic weight adjustments for both scenarios only apply to core risk data; the weights for other non-core data remain the same as the comprehensive weights, ultimately achieving a precise match between the data fusion weights and the risk characteristics of different mining scenarios.

[0232] For example, for "stress change data", its basic weight is calculated through comprehensive weighting. This value is applicable to general data fusion scenarios across the entire mine. The comprehensive weight is a basic fixed weight calculated by AHP combined with the entropy weight method, applicable to general scenarios across the entire mine. The weight adjustment function for tunneling and coal mining faces will amplify and adjust the comprehensive weight of the corresponding risk indicators in real time according to the on-site working conditions such as the distance of mining and tunneling advance, cycle pressure, and roof changes, distinguishing the differences in the importance of the core monitoring indicators of the two types of working faces. This is not just an example, but an essential dynamic optimization step in the entire weight calculation process.

[0233] In one embodiment, stress, microseismic, and gas monitoring points of underground mine monitoring equipment are used as nodes in a graph. The spatial distance between nodes is calculated based on the spatial coordinates of the monitoring points, and the edges of the graph are constructed based on the geological correlation between the monitoring points. Specifically, monitoring points within the same fault influence zone are considered to have the same geological correlation and are assigned a higher geological correlation coefficient.

[0234] Data from each monitoring point was preprocessed and features extracted. For stress monitoring points, features such as maximum principal stress (MPa), minimum principal stress (MPa), and stress concentration factor were extracted, with stress node features calculated using dynamic stress field model data. For microseismic monitoring points, features such as number of microseismic events (times), cumulative energy (J), and magnitude (dimensionless) were extracted, with microseismic node features calculated using real-time microseismic monitoring data. For gas monitoring points, features such as gas pressure (MPa), gas content (m³ / t), and gas concentration change rate (% / min) were extracted. For microseismic monitoring points, features such as number of microseismic events (times), cumulative energy (J), and magnitude were extracted. The extracted features for each node were standardized to facilitate the fusion and comparison of different data types.

[0235] In one embodiment, the graph neural network fusion model achieves synergistic use through a two-layer architecture of "feature layer fusion + weight layer optimization," forming a complementary fusion mechanism. This embodiment sets up two independent weight layers to achieve complementary fusion, with the two sets of weights participating in the data feature fusion calculation sequentially and in stages. The first layer consists of the bottom-level spatial edge weights within the graph neural network, while the second layer is an upper-level comprehensive dynamic weight constructed based on the AHP-entropy weight method and adapted to mining conditions. The two are completely distinct in their operational stages, target objects, and computational functions. Specifically, the order of computation is as follows:

[0236] (1) Input multi-source monitoring data and construct GNN graph nodes and edges;

[0237] (2) GNN internal message passing: Only the bottom-level spatial edge weights are used to complete the fusion of spatial correlation features between nodes;

[0238] (3) GNN outputs preliminary fused features;

[0239] (4) Introduce upper-level comprehensive dynamic weights (weights adjusted by the tunneling / coal mining condition function) to perform secondary weighted correction on the GNN output features;

[0240] (5) The corrected features are fed into the early warning judgment model to complete the risk classification.

[0241] The underlying spatial edge weights only participate in the computation during the message passing iteration of the graph neural network, and are used to characterize the spatial distance and the strength of the geological structural correlation between monitoring nodes. The edge weights between nodes comprehensively consider both the spatial distance and geological correlation between nodes, and the calculation formula is as follows:

[0242]

[0243] in, Let the weight of the edge between node i and node j be . This represents the spatial distance between node i and node j (in meters). This is the geological correlation coefficient, which increases when nodes i and j have the same geological correlation. This underlying weight is used to complete the spatial dimension information interaction and preliminary feature fusion of multiple underground measuring points, reflecting only the spatial geological coupling relationship between measuring points and not distinguishing between tunneling and coal mining conditions.

[0244] Data fusion is performed using the message passing mechanism of GNN. In each round of message passing, each node collects feature information from its neighboring nodes and updates its own features accordingly. Specifically, the feature update formula for node i is:

[0245]

[0246] in, This represents the feature vector of node i after the (t+1)th round of message passing (the dimension is determined by the number of features). This represents the feature vector of node i after the t-th round of message passing; This represents the edge weight between node i and node j; This represents the feature vector of node j after the t-th round of message passing; Represents the set of neighboring nodes of node i; This represents the activation function (such as the ReLU function). Activation functions are key functions in neural networks; their role is to introduce nonlinearity into the model, enabling it to fit complex nonlinear relationships. Taking the ReLU function as an example, its expression is: That is, when the input x is a positive number, the output is equal to x; when the input x is a negative number, the output is 0.

[0247] Through multiple rounds of message passing, nodes can fully absorb and integrate information from surrounding nodes, ensuring that each node's features contain multi-source data information from itself and related surrounding nodes, achieving deep integration of cross-type data. For example, a stress node located in a fault-affected area, after multiple rounds of message passing, will have its feature vector containing not only the stress data it monitors, but also information such as the displacement (unit: m) and activity rate (unit: mm / day) of adjacent fault nodes, as well as the gas pressure (unit: MPa) data of nearby gas nodes, thus more comprehensively reflecting the geological and mining conditions of the area.

[0248] The definition and secondary weighting of the upper-level weights are as follows: the upper-level comprehensive dynamic weights are obtained by weighting and fusing subjective weights obtained from the analytic hierarchy process (AHP) and objective weights obtained from the entropy weight method. They can also be dynamically corrected using adjustment functions based on field conditions such as the advance distance of the tunneling face, the cycle of the coal mining face, changes in the thickness of key roof layers, and fault activity rates. This is used to differentiate the contribution of different monitoring indicators, such as stress, microseismic activity, gas, and geological structure, to the early warning of dynamic disasters. It is important to note that these upper-level weights do not participate in the internal iterative calculations of the graph neural network. They are a post-weighting step after the graph neural network operation is completed. Only after the graph neural network outputs the fused features of all nodes, a secondary weighting correction is performed on the output features to strengthen the influence of key risk indicators under different mining scenarios.

[0249] For example, the stress eigenvector of the tunneling face. The final characteristic is ,in Adjusted by dynamic weight function Sure, As the initial weights, The adjustment factor is 0.5 in this embodiment, and L is the working face advance distance (unit: m). The total advancing length of the working face (unit: m).

[0250] Therefore, when describing the complete computational process of a graph neural network, only the underlying spatial edge weights used within that computational step are introduced. The upper-layer comprehensive dynamic weights are not internal computational parameters of the graph neural network, but are parameters for subsequent independent processing steps. The two layers of weights participate in the computation in stages: the underlying weights complete spatial correlation fusion, and the upper-layer dynamic weights complete adaptive correction of indicator importance. Together, they achieve refined weighted fusion of multi-source heterogeneous data.

[0251] In one embodiment, the multi-source data set (geostress field data) defined in the multi-data fusion mathematical model is used. Microseismic monitoring data Geological structural data Coal and rock mechanical property data (etc.) are used as node feature inputs for the GNN model. Among them, This is geostress field data (unit: MPa, megapascal), corresponding to the maximum principal stress, stress concentration factor, and other characteristics of stress nodes in the GNN; For microseismic monitoring data (unit: times / minute, used to characterize the frequency of microseismic events), the cumulative energy (unit: J, joules) and magnitude of microseismic events are used as evaluation indicators of GNN node activity. When the cumulative energy exceeds 5000J, the message passing weight of the corresponding node is increased by 20%, which enhances the influence of microseismic data in feature fusion. Geological structural data (such as fault drop ratios and normalized fold curvature values) are used to optimize the geological correlation coefficients between GNN nodes. (When the fault displacement ratio > 0.3,) (Increased from 1.5 to 1.8). The data pertains to the mechanical properties of coal (rock) (units vary depending on the specific parameters, such as elastic modulus in GPa and Poisson's ratio dimensionless). The elastic modulus and Poisson's ratio are used as correction terms in the GNN edge weight calculation. ,in The measured elastic modulus (unit: GPa). For reference values ​​(unit: GPa, usually taken as 30GPa).

[0252] In one embodiment, the collaborative optimization of dynamic weights is based on the actual needs of different scenarios such as coal mining faces and fault-affected areas. By establishing a dynamic weight collaborative optimization mechanism, the weight parameters are adaptively adjusted according to the real-time monitoring features output by the GNN and the weight calculation logic of the multi-data fusion mathematical model, thereby enhancing the model's dynamic response capability to complex working conditions.

[0253] In the context of coal mining faces, the periodic pressure features (such as microseismic energy abrupt change points) of nodes in the GNN are used as real-time correction values ​​(in meters) for the periodic pressure step size S in the multi-data fusion model. When the GNN detects a precursor to periodic pressure, the weight adjustment function is triggered. The advance calculation, of which Weighting of data related to the coal mining face. As the initial weights, This is the ratio of the actual advance distance to the step distance based on the cycle. This represents the variation in the thickness of the critical layer of the roof slab (unit: m). The initial thickness of the critical layer of the top slab (unit: m). This allows for weight adjustment 5-8 hours earlier than traditional methods;

[0254] In the context of a fault-affected zone, the fault activity rate output by the GNN is... (Unit: mm / day, millimeters / day) Geological structural data weights in multi-data fusion models Linkage, when mm / day, automatic lifting The value was increased to 1.3 times the initial value, strengthening the role of fault data in the fusion.

[0255] In one embodiment, the corrected stress value output by the GNN (Unit: MPa) The results of the comprehensive weight calculation of the multi-data fusion model are cross-validated: when the disaster risk assessment deviation rate of the two is <10%, it is judged as a high-confidence result; when the deviation rate is 10%-20%, supplementary monitoring is initiated (such as increasing the frequency of microseismic sampling); when the deviation rate is >20%, the model parameters are recalibrated to ensure the reliability of the fusion results.

[0256] In one embodiment, fused node features are used for dynamic disaster early warning. Fault nodes include elevation drop h (in meters) and activity rate. (Unit: mm / day, millimeters / day) and other data are transmitted to stress nodes within a 50m radius. These stress nodes, combining their own stress characteristics with information transmitted from fault nodes, correct the stress values.

[0257]

[0258] in, The measured stress values ​​at the stress nodes (unit: MPa, megapascals).

[0259] Gas nodes will put pressure (Unit: MPa) Stress nodes transmitted to the same coal seam, with stress concentration factors adjusted according to the magnitude of gas pressure:

[0260]

[0261] in, The original stress concentration factor is . This is the corrected stress concentration factor.

[0262] Based on the corrected stress values ​​and stress concentration factors, combined with preset warning thresholds, the system determines whether to issue a dynamic disaster warning signal. When the relevant indicators exceed the warning threshold, the system immediately issues a warning, prompting on-site personnel to take appropriate preventative measures to reduce the risk of dynamic disasters.

[0263] In one embodiment, to improve the model's prediction accuracy, a large model is iteratively trained using massive amounts of historical mining data. During training, the backpropagation algorithm and stochastic gradient descent method are used to optimize the model parameters. The backpropagation algorithm is used to calculate the error between the predicted and the true values ​​and propagates the error back from the output layer to the input layer, thereby adjusting the connection weights between layers; the stochastic gradient descent method calculates the gradient by randomly selecting a portion of the data to update the model parameters, avoiding getting trapped in local optima and thus reducing the model's false alarm rate.

[0264] In a preferred embodiment, a real-time self-optimization mechanism of reinforcement learning (RL) is introduced. The data from the monitoring points are used as the state, and model parameters (such as membership function thresholds, balance coefficients of the multi-data fusion model) are used as the parameters. The "action" is defined as "early warning accuracy - false alarm rate" as the reward function. The monitoring points are treated as the state of the intelligent agent, that is, the real-time data (such as stress value, micro-vibration frequency, gas pressure, dynamic weight, etc.) collected by the entire mine monitoring network (stress monitoring points, micro-seismic monitoring points, gas monitoring points, etc.) are integrated into the "current real-time working condition state" that the intelligent agent can recognize.

[0265] The intelligent agent iteratively adjusts parameters hourly. For example, after three consecutive false alarms, the alarm threshold for "microseismic energy release rate" is automatically increased from 100 J / s to 120 J / s, while the weight adjustment coefficient of stress data in the multi-data fusion model is increased from 0.5 to 0.7, improving model adaptability. The model has adaptive learning capabilities and can dynamically adjust the weights of risk characteristic parameters and model structural parameters based on real-time data feedback, continuously enhancing the reliability and timeliness of early warnings.

[0266] In one embodiment, based on the results of multi-data fusion, a mathematical model is constructed using fuzzy comprehensive evaluation and analytic hierarchy process (AHP) to quantitatively classify dynamic hazards in deep coal mines, dividing the early warning levels into four levels: safe (Level I), caution (Level II), early warning (Level III), and alarm (Level IV). The fused geostress field data, dynamic hazard sensitive index data, microseismic monitoring data, and geological structure data constitute the index set. . Maximum principal stress (MPa) reflects the maximum stress value experienced by the coal and rock mass; Stress concentration factor (dimensionless), the ratio of measured stress to original rock stress; The coal impact energy index (KE) characterizes the coal's tendency to be impacted. : Frequency of microseismic events (times / day), the number of microseismic events per unit time; Microseismic energy release rate (J / s) reflects the intensity of energy release from the coal and rock mass; Fault influence coefficient (dimensionless), a parameter related to fault distance and elevation difference. Warning level set. These correspond to safety, caution, early warning, and alarm.

[0267] The weights of each indicator are determined using the analytic hierarchy process (AHP), and a judgment matrix is ​​constructed. ,in Indicators relative to indicators The importance ratio. The weight vector is obtained by calculating the matrix eigenvectors and the largest eigenvalue. ,in Typical weight allocation example: (Maximum principal stress): 0.25; (Stress concentration factor): 0.20; (Impact Energy Index): 0.15; (Micro-vibration frequency): 0.15; (Microseismic energy release rate): 0.15; (Fault influence coefficient): 0.10. The weight vector represents the relative importance of each early warning indicator in the dynamic disaster rating.

[0268] Establish membership function Quantify the degree of membership of each indicator to different warning levels to form a membership matrix. , Indicators For the warning level Membership degree. Based on the maximum principal stress. For example, the membership function takes the form:

[0269] .

[0270] Therefore, the comprehensive membership vector is calculated through fuzzy transformation. , ,in Indicates the overall evaluation of the grade. The degree of membership.

[0271] According to the principle of maximum membership, if If so, the current dynamic disaster warning level is determined to be... .

[0272] In practical applications, historical monitoring data and disaster cases are used to train the model, optimize the membership function and weight vector, and dynamically adjust the early warning threshold. For example:

[0273] Safety thresholds: maximum principal stress < 15 MPa, impact energy index < 5;

[0274] Note the thresholds: 15MPa≤maximum principal stress<25MPa, 5≤impact energy index<10;

[0275] Warning thresholds: 25MPa≤maximum principal stress<35MPa, 10≤impact energy index<15;

[0276] Alarm threshold: maximum principal stress ≥35MPa, impact energy index ≥15.

[0277] As a preferred implementation method, the data weights are adjusted by combining the characteristics of tunneling and coal mining faces. For example, the stress change index (weight 0.30) is given a higher weight in tunneling faces, and the roof and floor stress index (weight 0.35) is given a higher weight in coal mining faces, so as to achieve accurate early warning and provide a scientific basis for safe production in deep coal mines.

[0278] like Figure 2 As shown, this embodiment of the invention also provides a coal mine large-scale model dynamic disaster early warning system based on stress field analysis and multi-data source fusion, including:

[0279] The initial modeling module is used to acquire and integrate multi-source data including mine geological parameters, coal and rock mechanics parameters, and measured underground geostress parameters, and to construct an initial geostress field model through numerical simulation.

[0280] The indicator evaluation module is used to establish a dynamic disaster sensitivity indicator evaluation system based on multi-source data and an initial geostress field model. This system includes four dimensions of indicators: stress state, geological structure, coal and rock characteristics, and gas-related indicators. The analytic hierarchy process is used to determine the weight of each indicator and to construct a comprehensive evaluation function for risk classification.

[0281] The dynamic correction module is used to acquire real-time data from detection equipment deployed at the tunneling face, coal mining face, and mining area, compare and analyze the real-time data with the initial geostress field model, construct a difference function to determine the dynamic correction coefficient, and dynamically correct the initial geostress field model according to the dynamic correction coefficient of each area to generate a dynamic stress field model.

[0282] The early warning output module is used to construct a graph neural network fusion model. It takes the dynamic stress field model, risk classification results and real-time data as input to the graph neural network fusion model, uses a message passing mechanism to fuse the input data, and introduces upper-level comprehensive dynamic weights to perform weighted optimization on the fused data to obtain the final fused features. The final fused features are then input into a preset early warning level determination model to output the disaster early warning level.

[0283] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early warning of dynamic disasters in coal mines based on stress field analysis and multi-data source fusion, characterized in that, include: Step 1: Acquire and integrate multi-source data including mine geological parameters, coal and rock mechanics parameters, and measured underground in-situ stress parameters, and construct an initial in-situ stress field model using numerical simulation. Step 2: Based on the multi-source data and the initial geostress field model, establish a dynamic disaster sensitivity index evaluation system that includes four dimensions of indicators: stress state, geological structure, coal and rock characteristics, and gas-related indicators. Use the analytic hierarchy process (AHP) to determine the weight of each indicator and construct a comprehensive evaluation function for risk classification. Step 3: By acquiring real-time data from detection equipment deployed at the tunneling face, coal mining face, and mining area, the real-time data is compared and analyzed with the initial geostress field model to construct a difference function to determine the dynamic correction coefficient. The initial geostress field model is then dynamically corrected based on the dynamic correction coefficient for each region to generate a dynamic stress field model. Step 4: Construct a graph neural network fusion model. The dynamic stress field model, risk classification results, and real-time data are used as inputs to the graph neural network fusion model. The input data is fused using a message passing mechanism, and upper-level comprehensive dynamic weights are introduced to perform weighted optimization on the fused data to obtain the final fused features. The final fused features are then input into a preset early warning level determination model to output the disaster early warning level.

2. The method for early warning of dynamic disasters in large coal mine models based on stress field analysis and multi-data source fusion as described in claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Obtain mine geological parameters and coal and rock mechanics parameters, specifically including: obtaining data on the thickness and occurrence of strata lithology; obtaining the curvature radius parameters of synclines and anticlines of fold structures; obtaining the strike, dip angle, length, and displacement parameters of faults, and calculating fault activity rate parameters based on microseismic monitoring data; and constructing normal fault fusion models and reverse fault fusion models based on the fault activity rate parameters to correct the geostress. Step 1.2: Obtain measured parameters of downhole in-situ stress, including: measured data of triaxial principal stress, stress gradient and six-axis stress components of rock mass obtained by hydraulic fracturing method or stress relief method; Step 1.3: Obtain the correlation characteristic data of geostress direction, stress anomaly and tectonic stress energy, including: microseismic event location parameters and source energy parameters obtained from microseismic monitoring data inversion, and seismic wave velocity anomaly parameters and reflected wave imaging parameters obtained from 3D seismic exploration data inversion; Step 1.4: Using the mine geological parameters, coal and rock mechanical parameters, and tectonic stress energy correlation characteristic data as input, perform simulation calculations using finite element or discrete element numerical models to generate simulation results of the global initial geostress field; Step 1.5: Compare and verify the simulation results of the global initial geostress field with the measured parameters of the downhole geostress field and iteratively correct them until the simulation result error is lower than the preset threshold, and generate the initial geostress field model.

3. The method for early warning of dynamic disasters in large coal mine models based on stress field analysis and multi-data source fusion as described in claim 1, characterized in that, In step 2, the stress state indicators include the maximum principal stress concentration factor, stress gradient, and stress-gas coupling hazard index; the geological structure indicators include the fault influence factor and fold curvature index; the coal and rock properties indicators include the coal impact tendency index and rock brittleness index; and the gas-related indicators include gas pressure, gas content, gas dissipation index, borehole gas emission initial velocity, drill cuttings volume, and rock damage evolution coefficient.

4. The method for early warning of dynamic disasters in large coal mine models based on stress field analysis and multi-data source fusion as described in claim 1, characterized in that, In step 3, the formula for the difference function is: in, This represents the stress change vector between real-time data and the initial geostress field model. Assigning weights to stress data in each direction. This represents the stress vector of the initial geostress field model. This represents the geostress field vector of real-time data; Based on the aforementioned difference function, the stress field is updated using an adaptive correction algorithm, and the correction formula is as follows: in, This represents the stress vector in a long-range dynamic stress model. This represents the correction factor. This represents the stress variation vector.

5. The method for early warning of dynamic disasters in large coal mine models based on stress field analysis and multi-data source fusion as described in claim 1, characterized in that, Step 4, the construction of the graph neural network fusion model, specifically includes: Using stress, microseismic, and gas monitoring points of underground mine monitoring equipment as nodes of a graph neural network, the spatial distance between nodes is calculated based on the spatial coordinates of the monitoring points, and the edges of the graph are constructed based on the geological correlation between the monitoring points. Calculate the feature vector of each node; wherein the features of the stress node are calculated using the dynamic stress field model data, the features of the microseismic node are calculated using real-time microseismic monitoring data, the features of the gas node are calculated using the gas monitoring data, and the features of each node are integrated with the index data in the sensitive index evaluation system. Based on the edge weights, each node aggregates the feature information of its neighboring nodes through the message passing mechanism of the graph neural network, and updates the node features to complete the input feature fusion: in, This represents the feature vector of node i after the (t+1)th round of message passing; This represents the feature vector of node i after the t-th round of message passing; This represents the edge weight between node i and node j; This represents the feature vector of node j after the t-th round of message passing; Represents the set of neighboring nodes of node i; Indicates the activation function; A second weighting of the fused features is applied by introducing upper-level comprehensive dynamic weights to obtain the final fused features.

6. The method for early warning of dynamic disasters in large coal mine models based on stress field analysis and multi-data source fusion as described in claim 5, characterized in that, The upper-layer integrated dynamic weights perform secondary weighted optimization on the output features after the graph neural network fusion model outputs node fusion features, specifically including: in, This represents the overall weight of the k-th type of node. Represents the balance coefficient. This represents the objective weights calculated using the entropy weight method. This represents the subjective weights determined using the analytic hierarchy process (AHP).

7. The method for early warning of dynamic disasters in large coal mine models based on stress field analysis and multi-data source fusion as described in claim 1, characterized in that, The process of inputting the fused features into the early warning level determination model to obtain the dynamic disaster early warning result specifically includes: The features output by the graph neural network fusion model are mapped to a set of early warning judgment indicators; Establish the membership function of each indicator in the early warning judgment indicator set to different early warning levels, and form a membership matrix; The dynamic weight vector of the index is combined with the membership matrix by a fuzzy synthesis operation to obtain the comprehensive membership vector. Based on the principle of maximum membership, the risk level of dynamic disasters is determined from the comprehensive membership vector, and a dynamic disaster early warning signal is output.

8. The method for early warning of dynamic disasters in large coal mine models based on stress field analysis and multi-data source fusion as described in claim 1, characterized in that, Step 4 further includes: introducing a reinforcement learning real-time self-optimization mechanism, taking the monitoring points as the state of the agent, taking the parameters of the graph neural network fusion model as the action of the agent, taking the comprehensive index of warning accuracy and false alarm rate as the reward function, and adjusting the parameters of the graph neural network fusion model iteratively through the agent to achieve adaptive optimization of the graph neural network fusion model.

9. A coal mine large-scale model dynamic disaster early warning system based on stress field analysis and multi-data source fusion, characterized in that, include: The initial modeling module is used to acquire and integrate multi-source data including mine geological parameters, coal and rock mechanics parameters, and measured underground geostress parameters, and to construct an initial geostress field model through numerical simulation. The indicator evaluation module is used to establish a dynamic disaster sensitivity indicator evaluation system based on the multi-source data and the initial geostress field model. This system includes four dimensions of indicators: stress state, geological structure, coal and rock characteristics, and gas-related indicators. The weight of each indicator is determined by the analytic hierarchy process (AHP), and a comprehensive evaluation function is constructed for risk classification. The dynamic correction module is used to acquire real-time data from detection equipment deployed at the tunneling face, coal mining face, and mining area, compare and analyze the real-time data with the initial geostress field model, construct a difference function to determine the dynamic correction coefficient, and dynamically correct the initial geostress field model according to the dynamic correction coefficient of each region to generate a dynamic stress field model. The early warning output module is used to construct a graph neural network fusion model. The dynamic stress field model, risk classification results, and real-time data are used as inputs to the graph neural network fusion model. The input data is fused using a message passing mechanism, and upper-level comprehensive dynamic weights are introduced to perform weighted optimization on the fused data to obtain the final fused features. The final fused features are then input into a preset early warning level determination model to output the disaster early warning level.