Quantitative characterization method for outburst danger of deep coal seam based on multi-field coupling effect

Abstract

The present application relates to the field of deep coal mining disaster prevention and control, and relates to a deep coal seam outburst risk quantitative characterization method based on multi-field coupling effect, comprising: S1, establishing a codeable heat-flow-solid-coupling four-field coupling control equation set; S2, constructing a chemical damage evolution model; S3, based on the field variable distribution obtained by solving the control equation set, establishing an energy catastrophe instability criterion with the system total potential energy second-order variation as the core, and determining the critical energy threshold of outburst danger; S4, constructing and training a physical model driven neural network to realize rapid prediction of field variables; S5, using the trained physical model driven neural network for rapid prediction, and realizing quantitative grading and engineering early warning of deep coal seam outburst risk based on comparison of the prediction result with the critical energy threshold. The present application provides scientific, reliable and practical technical support for precise prevention and control of deep coal mining outburst disasters.

Description

Technical Field

[0001] This invention relates to the field of disaster prevention and control technology in deep coal mining, specifically to a quantitative characterization method for the outburst risk of deep coal seams based on the multi-field coupling effect of Thermal-Hydraulic-Mechanical-Chemical (THMC). This method is applicable to the accurate evaluation and real-time prediction of coal and gas outburst risks in deep coal seams under conditions of high ground temperature, high ground stress, and high gas pressure. Background Technology

[0002] Coal and gas outbursts are one of the most serious geological hazards in deep coal mining, directly threatening mine safety and the lives of workers. As the depth of coal mining continues to increase, the geological conditions of coal seams become increasingly complex. The "three highs" characteristics of high ground temperature, high ground stress, and high gas pressure make the multi-field coupling effect of THMC (thermal, thermal, and gas) more and more significant, becoming the core disaster-causing factor that induces coal and gas outbursts. Current methods for assessing the risk of outbursts in deep coal seams still suffer from several technical shortcomings: First, there is a logical disconnect between the coupling of the THMC physical model and the artificial intelligence (AI) algorithm. The physical model lacks a closed and coded set of control equations, and the AI ​​algorithm lacks physical constraints, leading to a disconnect between the prediction results and engineering realities. Second, the core control equations are mostly conceptual frameworks, and the parameter calibration methods are unclear. Key parameters are difficult to obtain during engineering implementation, resulting in poor engineering operability of the model. Third, multi-field coupled numerical simulations are computationally complex and inefficient, failing to meet the engineering requirements for real-time on-site prediction. Fourth, outburst risk assessments are mostly based on qualitative analysis, and quantitative criteria lack clear physical meaning. The universality of critical thresholds is poor, making it difficult to adapt to spatial variations in geological conditions. Fifth, the characterization of field variables is prone to information loss, failing to effectively capture early signs of instability such as localized coal seam damage, resulting in insufficient accuracy of the assessment results.

[0003] In existing technologies, some methods only consider the coupling effect between fluid and solid fields, neglecting the regulatory role of temperature and chemical fields on coal seam outbursts, and thus failing to reflect the disaster-causing mechanisms of deep "three-high" environments (high temperature, high humidity, and high gas content). Some methods introduce AI algorithms for outburst prediction, but rely solely on data-driven approaches without considering the physical laws of multi-field coupling in THMC (Thermal, Hydrogen, and Methane) systems, resulting in poor model generalization ability and low training stability. Other methods construct multi-field coupling models but lack efficient surrogate models, leading to low computational efficiency of traditional numerical simulation methods and difficulty in real-time field applications. Furthermore, existing methods often rely on single indicators for outburst risk assessment, lacking global criteria based on system energy evolution, and thus failing to accurately characterize the abrupt transition from steady-state to unsteady-state processes in the coal-gas system. In summary, existing technologies struggle to achieve high-precision, quantitative, and real-time evaluation of deep coal seam outburst risks. There is an urgent need to develop a logically closed-loop, highly implementable, theoretically rigorous, and engineering-practical quantitative characterization method for deep coal seam outburst risks, addressing core technical challenges such as the coupling of physical models and AI algorithms, model engineering implementation, efficient prediction, and quantitative criteria. Summary of the Invention

[0004] To address the shortcomings of existing methods for assessing the risk of outbursts in deep coal seams, such as logical gaps in the coupling between the THMC physical model and the AI ​​algorithm, lack of closed and coded forms in the core control equations, difficulty in obtaining and calibrating engineering implementation parameters, disconnect between AI prediction targets and the evolution logic of field variables, loss of characteristic representation information of field variables, poor universality of critical thresholds, and low computational efficiency, this invention provides a quantitative characterization method for the risk of outbursts in deep coal seams based on the multi-field coupling effect of THMC. The purpose of this invention is to: construct a complete and coded four-field coupled control equation set for THMC, determine the calibration and acquisition methods for all parameters, and improve the engineering feasibility of the physical model; define a chemical damage evolution model to quantitatively characterize the deteriorating effect of chemical fields on coal strength and improve the strong coupling mechanism of THMC multi-fields; derive an energy mutation instability criterion based on global integration of field variables, establish a quantitative evaluation standard with clear physical meaning, and solve the problem of qualitative evaluation in traditional evaluation methods; train a physical model-driven neural network (PMNN) adapted to integer-order THMC systems to achieve deep coupling between the physical model and AI algorithms, and solve the pain point of low computational efficiency in multi-field coupled simulation; supplement key details such as feature selection, numerical format implementation, and critical threshold adaptability during implementation, forming a complete technical system from theoretical modeling, numerical solution to surrogate model training and engineering application, realizing high-precision theoretical modeling, low-cost numerical solution, and rapid engineering prediction of outburst risk in deep coal seams, and providing scientific, reliable, and practical technical support for the precise prevention and control of outburst disasters in deep coal mining.

[0005] To achieve the above objectives, the following technical solution is adopted:

[0006] This invention provides a quantitative characterization method for the outburst risk of deep coal seams based on multi-field coupling effects, comprising the following steps: S1, establishing an coded set of thermal-fluid-solid-chemical four-field coupling control equations. The control equations are based on the occurrence characteristics of deep coal seams, and the effective stress equation is modified by introducing temperature stress terms and chemisorption stress terms. A permeability dynamic evolution equation including chemical damage variables is constructed to fully describe the strong coupling mechanism between the temperature field, seepage field, stress field, and chemical field; S2, constructing a chemical damage evolution model based on the chemical potential gradient. By defining chemical damage variables and their deterioration relationship with coal body mechanical parameters, the degree of damage to coal body strength by the chemical field is quantitatively characterized; S3, based on the field variable distribution obtained by solving the control equations, the coal body's elastic-plastic deformation energy and internal gas energy are calculated by globally integrating them. - The total potential energy of the gas system is used to establish an energy mutation instability criterion with the second-order variation of the total potential energy as the core, and to determine the critical energy threshold for outburst risk; S4. A physical model-driven neural network is constructed and trained. This network uses a high-order time discrete iterative function derived based on the control equations as physical constraints to learn the spatiotemporal mapping law from the field variables and boundary conditions at the current time to the field variables at the next time, serving as a surrogate model for the control equations to achieve rapid prediction of field variables; S5. By comprehensively applying the above steps S1-S4, parameters are calibrated through laboratory experiments, field data is collected, critical thresholds are solved through numerical simulation, and rapid prediction is performed using the trained physical model-driven neural network. Based on the comparison between the prediction results and the critical energy threshold, the quantitative classification and engineering early warning of outburst risk in deep coal seams are finally achieved.

[0007] Furthermore, in step S1, a set of coded thermal-fluid-solid-chemical four-field coupled control equations is established, specifically including: constructing a constitutive equation for the total strain of the coal body: the total strain of the coal body consists of four parts: elastic strain, plastic strain, thermal expansion strain, and gas adsorption expansion strain, to describe the deformation behavior of the coal body under multi-field coupling; constructing a modified effective stress equation: adding temperature stress terms and chemical adsorption stress terms to the traditional effective stress to quantitatively characterize the regulation mechanism of effective stress of the coal body by temperature changes and gas adsorption / desorption; constructing a dynamic evolution equation for permeability: by coupling the effects of coal body volumetric strain, chemical damage variables, and temperature changes, the dynamic updating of coal seam permeability during mining disturbance is realized; constructing a constitutive equation for gas adsorption considering the influence of temperature: by introducing the correction of Langmuir adsorption parameters by temperature, the variation law of gas adsorption / desorption behavior under high geothermal environment is accurately described.

[0008] Furthermore, the construction of the chemical damage evolution model in step S2 specifically includes: defining chemical damage variables based on the chemical potential gradient, wherein the chemical damage variables are used to quantitatively characterize the degree of deterioration of the coal body caused by the initiation and propagation of microcracks induced by coal matrix shrinkage during gas desorption, and their values ​​range between no damage and complete destruction; calibrating the chemical damage coefficient by conducting adsorption / desorption experiments on coal samples under different gas pressures, simultaneously monitoring acoustic emission signals to characterize the degree of microcrack development, fitting the relationship between the degree of damage and the chemical potential gradient, and obtaining the calibrated value of the chemical damage coefficient; and establishing the dynamic evolution relationship of the chemical damage variables by substituting the change law of gas desorption over time into the chemical potential gradient calculation formula to realize the dynamic updating of the chemical damage variables with the mining disturbance time.

[0009] Furthermore, step S2 also includes: constructing a coupling relationship between chemical damage and coal strength, introducing the chemical damage variable into the strength parameters of the coal, and modifying the original cohesion and original internal friction angle of the coal to achieve the coupling effect of the chemical field on the stress field. The modification is achieved through the combination relationship between the chemical damage variable and the strength degradation coefficient.

[0010] Furthermore, in step S3, the calculation of the total potential energy of the coal-gas system by globally integrating the elastic-plastic deformation energy of the coal body and the internal energy of the gas includes: calculating the total potential energy of the system composed of the elastic-plastic deformation energy of the coal body and the internal energy of the gas by globally integrating the study area; wherein, the elastic-plastic deformation energy of the coal body is calculated based on the stress field and the elastic strain field, and only the elastic strain part is included, while the plastic deformation part is not included as dissipated energy; the internal energy of the gas is calculated based on the gas pressure field, temperature field, and gas physical property parameters, according to the ideal gas law and the thermodynamic internal energy formula.

[0011] Furthermore, in step S3, establishing an energy abrupt change instability criterion centered on the second-order variation of the system's total potential energy, and determining the critical energy threshold for outburst danger, specifically includes: using the second-order variation of the system's total potential energy as a quantitative basis for determining the stability of the coal-gas system; when the second-order variation is greater than 0, the system is determined to be in a steady state, and there is no outburst danger in the coal seam; when the second-order variation is equal to 0, the system is determined to be in a critical instability state, and the total potential energy of the system at this time is determined as the critical energy threshold for coal seam outburst danger; when the second-order variation is less than 0, the system is determined to be in an unsteady state, and a coal and gas outburst will occur in the coal seam.

[0012] Furthermore, in step S4, the constructed and trained physical model-driven neural network is positioned as a surrogate model for the four-field coupled control equations of heat-fluid-solid-chemical system. Specifically, it includes: constructing a time iteration scheme for the integer-order heat-fluid-solid-chemical system based on a high-order time discrete format, using the spatiotemporal mapping law implied by this iteration scheme as the learning target of the neural network; the high-order time discrete format adopts... The format discretizes the time derivative terms in the governing equations and constructs an iterative function from the distribution of field variables and boundary conditions at the previous time step to the distribution of field variables at the next time step. The physical model drives the neural network to learn the input-output mapping relationship of the iterative function, replacing the step-by-step solution process of traditional numerical simulation, and realizing the rapid prediction of field variables.

[0013] Furthermore, the physical model-driven neural network adopts an end-to-end network structure comprising an input layer, a hidden layer, a spatiotemporal mapping layer, and an output layer. The input layer receives the field variable features and boundary conditions at the current time step. These field variable features simultaneously include macroscopic statistical features describing the overall level of the field distribution and spatial variation features for capturing instability precursors such as damage localization. The hidden layer consists of multiple fully connected neurons for nonlinear feature extraction and mapping. The spatiotemporal mapping layer employs a convolutional structure to maintain the iterative continuity of the field variables over time. The output layer outputs the field variable features for the next time step, with the same dimension as the field variable features of the input layer.

[0014] Furthermore, in step S4, constructing and training the physical model-driven neural network also includes the following steps: generating training data using a hybrid data strategy combining numerical simulation samples and field-measured samples, wherein the main training set is generated by thermal-fluid-solid-chemical numerical simulation, and the calibration set is obtained by inversion from field-measured data; constructing a composite loss function including field variable fitting loss and physical iteration constraint loss, wherein the physical iteration constraint loss is constructed based on the iteration function and is used to constrain the neural network prediction results to conform to the physical evolution law of the thermal-fluid-solid-chemical system; training the network using an optimization algorithm, and fine-tuning the parameters of the trained model using the calibration set, so that the error between the model prediction results and the field-measured data is controlled within a preset range; using the trained and calibrated physical model-driven neural network for rapid on-site prediction, and using the field variable characteristics of its prediction output as input for subsequent energy mutation instability criterion calculation and hazard classification determination, forming a two-step engineering application mode of rapid prediction-quantitative determination.

[0015] Furthermore, step S5 includes: inputting the field variable characteristics of the current moment monitored in real time into the trained and calibrated physical model to drive the neural network, predicting the field variable distribution at the next moment; selecting the corresponding critical energy threshold according to the geological sub-region to which the predicted location belongs, or estimating the critical energy threshold through an association model, and calculating the ratio of the total potential energy of the real-time system to the critical energy threshold; determining the outburst risk level of the coal seam based on the ratio according to the preset quantitative grading standard, and formulating corresponding disaster prevention and control measures accordingly to achieve accurate early warning and prevention of outburst disasters.

[0016] Compared with existing technologies, this invention, by constructing a THMC multi-field coupled control system, a chemical damage evolution model, an energy mutation instability criterion, and a physical driving proxy model, forms a logically closed-loop quantitative characterization method for the outburst risk of deep coal seams. This method has significant effects and advantages, as detailed below:

[0017] 1. Clarify the core positioning of PMNN as a proxy model for integer-order THMC systems, based on A higher-order time discrete scheme is used to construct an iterative scheme, embedding the iterative laws of the physical model into the neural network loss function. This enables PMNN prediction results to conform to physical laws and possess rapid spatiotemporal mapping capabilities. At the same time, the prediction of field variable features and the determination of energy criteria are implemented in separate modules, solving the technical problem that global integral quantities are difficult to predict directly, and realizing a reasonable and deep integration of multi-field coupling models and AI algorithms.

[0018] 2. For the first time, the temperature stress term was included. Chemical adsorption stress term By incorporating the effective stress equation and coupling the combined effects of temperature field and chemical field on permeability and coal strength, a complete set of strongly coupled four-field THMC control equations was constructed. All equations are given in a coded form, which fully reflects the multi-field coupling disaster mechanism in the deep "three-high" environment. Compared with the traditional fluid-solid coupling model, it is more in line with engineering practice.

[0019] 3. Taking the energy evolution law of the coal-gas system as the core, derive the energy abrupt instability criterion based on the global integral of field variables, and apply the second variation of the system's total potential energy. As a criterion for stability determination, The ratio is used to establish a four-level quantitative grading standard. The evaluation index has a clear physical meaning and can accurately characterize the sudden change process of coal from steady state to unsteady state. Compared with the traditional single-index qualitative evaluation, the accuracy and reliability are greatly improved.

[0020] 4. A lightweight PMNN network structure of "fully connected layer + spatiotemporal mapping layer" is designed. A hybrid data strategy of "simplified numerical simulation samples + field-measured calibration samples" is adopted to reduce the model training cost. The single-step prediction calculation cost of PMNN is <1s, which is more than 100 times more efficient than traditional THMC numerical simulation calculation. After field calibration, the prediction error is <5%. While ensuring high accuracy, it realizes real-time prediction in the field and solves the engineering pain point of low efficiency of traditional simulation methods.

[0021] 5. The laboratory calibration methods and engineering inversion approaches for all model parameters were clearly defined, and the chemical potential gradient was resolved. Chemical damage coefficient The problem of obtaining key parameters is addressed by using a combination of geological zoning and geological parameter correlation models to achieve the critical energy threshold. The spatial dynamic update solves the problem of poor universality of critical thresholds; it introduces a field variable representation method of "macro-statistical features + spatial variation features", and selects core features through feature importance analysis to effectively capture instability precursors such as damage localization and avoid information loss; all models can be further developed in commercial simulation software, and PMNN can be deployed on embedded terminals without the need for independent development of simulation platforms, making it highly practical for engineering.

[0022] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0023] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the invention. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0024] Figure 1 This is a flowchart illustrating the quantitative characterization method for outburst risk in deep coal seams based on multi-field coupling effect, according to an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the dynamic evolution of chemical damage variables with mining disturbance time in an embodiment of the present invention;

[0026] Figure 3 This is a schematic diagram of the curve showing the change of the total potential energy of the system with the mining process in an embodiment of the present invention;

[0027] Figure 4 This is a schematic diagram of the PMNN network structure in an embodiment of the present invention;

[0028] Figure 5This is a comparison of the training loss curves of the algorithm of this invention with those of different existing algorithms. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0031] To address the shortcomings of existing methods for assessing the risk of outbursts in deep coal seams, such as logical gaps in the coupling between the Thermal-Hydraulic-Mechanical-Chemical (THMC) physical model and the Artificial Intelligence (AI) algorithm, core control equations being merely conceptual frameworks lacking closed, coded forms, difficulty in obtaining and calibrating key parameters for engineering implementation, disconnect between AI prediction targets and the evolution logic of field variables, and the potential for information loss in the characterization of field variables and the inability to adapt critical thresholds to changes in geological conditions during implementation, this invention provides a logically closed-loop and highly implementable quantitative characterization method for the risk of outbursts in deep coal seams based on the multi-field coupling effect of THMC. This method achieves high-precision theoretical modeling, low-cost numerical solution, and rapid engineering prediction of coal seam outburst risk under deep high geothermal and high geostress environments by constructing a complete and coded four-field coupled control equation set of THMC, defining a chemical damage evolution model, deriving an energy mutation instability criterion based on global integration of field variables, and training a Physical Model-driven Neural Network (PMNN) adapted to integer-order THMC systems. It also addresses the technical pain points of low computational efficiency of multi-field coupled models and lack of physical constraints in AI algorithms. Furthermore, it supplements key optimization ideas for feature selection, numerical format implementation, and critical threshold adaptability during implementation, further enhancing the practicality of the solution for engineering application. This provides a technical solution for the prevention and control of outburst disasters in deep coal mining that combines theoretical rigor with engineering operability.

[0032] This invention presents a quantitative characterization method for outburst risk in deep coal seams based on the multi-field coupling effect of the Thermochemical Modeling (THMC). Following the modern engineering research paradigm of theoretical modeling, numerical solution, surrogate model training, and engineering application, the method employs five core steps: establishing an encodeable set of four-field coupled THMC control equations; constructing a chemical damage evolution model; deriving an energy mutation instability criterion based on global integration of field variables; training a PMNN adapted to an integer-order THMC system for rapid prediction; and implementing comprehensive engineering applications. This method achieves a quantitative characterization of outburst risk in deep coal seams. Each step is logically closed-loop and tightly coupled, with all terms standardized and symbols clearly defined. The specific technical solution is as follows:

[0033] Figure 1 This is a schematic flowchart of a method for quantitatively characterizing the outburst risk of deep coal seams based on multi-field coupling effects, according to an embodiment of the present invention. Figure 1 As shown, a quantitative characterization method for the outburst risk of deep coal seams based on multi-field coupling effects includes the following steps:

[0034] S1. Establish a set of coded thermal-fluid-solid-chemical four-field coupled control equations. The control equations are based on the occurrence characteristics of deep coal seams. The effective stress equation is modified by introducing temperature stress terms and chemical adsorption stress terms. A permeability dynamic evolution equation containing chemical damage variables is constructed to fully describe the strong coupling mechanism between temperature field, seepage field, stress field and chemical field.

[0035] Step S1 is used to establish a coded THMC four-field coupled control equation set.

[0036] Based on the occurrence characteristics of deep coal seams, namely "high ground stress, high gas pressure, and high ground temperature," and combining classical theories of coal rock mechanics, gas seepage, heat conduction, and surface physical chemistry, a set of four-field coupled control equations of THMC (Heat-Metal-Chemistry) is constructed, which includes specific constitutive forms, calibrable parameters, and closable equations. The mutual coupling law of temperature field (T), seepage field (H), stress field (M), and chemical field (C) is clarified. All equations are given specific forms that can be directly encoded and implemented, and the physical meaning and acquisition method of parameters are clarified. The core modification idea is to introduce the regulatory effect of temperature field on gas adsorption and coal deformation, and the deterioration effect of chemical field on coal damage and permeability evolution on the basis of classical fluid-solid (HM) coupling model, so as to achieve strong coupling of the four fields.

[0037] S11: Constructing the total strain constitutive equation for the coal body

[0038] Constructing the constitutive equation for total coal strain: The total coal strain consists of four parts: elastic strain, plastic strain, thermal expansion strain, and gas adsorption expansion strain, to describe the deformation behavior of the coal under multi-field coupling. Specifically, it can be coded as follows:

[0039]

[0040] In the formula, Let be the total strain tensor of the coal body. For elastic strain tensor, For plastic strain tensor, For thermal expansion strain tensor, Let S be the gas adsorption expansion strain tensor. , These are spatial coordinate components, with values ​​of 1, 2, and 3, representing the three coordinate directions in three-dimensional space (such as the x, y, and z axes).

[0041] The specific expressions and parameter definitions for each strain component are as follows:

[0042] (1) Elastic strain: In the formula, The shear modulus of the coal mass is determined by laboratory uniaxial / triaxial compression tests. Poisson's ratio, calibrated by laboratory uniaxial / triaxial compression experiments; This represents the total stress tensor of the coal body. Let be the volumetric stress, and be the trace of the total stress tensor; The symbol for Kronecker. hour , hour ;

[0043] (2) Plastic strain: The Mohr-Coulomb yield criterion is used to solve for it. In the formula As a plastic multiplier, The Mohr-Coulomb yield function is determined by classical formulas in coal and rock mechanics.

[0044] (3) Thermal expansion strain: In the formula The linear thermal expansion coefficient of the coal body is determined by laboratory thermal expansion experiments. This represents the real-time temperature of the coal mass, in units of... ; Original ground temperature, unit: The data was obtained from on-site geological exploration and measurement.

[0045] (4) Adsorption expansion strain: In the formula The limiting adsorption expansion strain of the coal body is determined by laboratory gas adsorption expansion experiments. Langmuir adsorption constant, in units of The adsorption was calibrated by laboratory volumetric adsorption experiments. Gas pore pressure, unit: .

[0046] S12: Construct the modified effective stress equation

[0047] A revised effective stress equation is constructed: Breaking through the limitation of traditional effective stress only considering pore gas pressure, a specific form of temperature stress term and chemical adsorption stress term are added to the traditional effective stress equation to quantitatively characterize the regulatory mechanism of temperature change and gas adsorption / desorption on the effective stress of the coal body. The specific coded form is as follows:

[0048]

[0049] In the formula, The effective stress tensor of the coal body; The effective stress coefficient; This is the temperature stress term, in units of... This is the chemical adsorption stress term, in units of... ; where i and j are spatial coordinate component indices, with values ​​of 1, 2, and 3, representing the three coordinate directions in three-dimensional space (such as the x, y, and z axes).

[0050] The specific expressions and parameter definitions for each stress term are as follows:

[0051] (1) Effective stress coefficient : Take coal porosity ,Right now Dimensionless, determined by laboratory mercury porosimetry.

[0052] (2) Temperature stress term : In the formula The elastic modulus of coal, in units of Calibrated by laboratory uniaxial compression experiments;

[0053] (3) Chemical adsorption stress term : The stress generated by gas adsorption expansion / desorption contraction is quantitatively characterized by the transformation of adsorption expansion strain through Hooke's Law.

[0054] S13: Constructing a dynamic evolution equation for penetration rate

[0055] A dynamic permeability evolution equation is constructed: By coupling the comprehensive effects of coal volumetric strain, chemical damage variables, and temperature changes on permeability, and using the classical cubic law of coal and rock damage mechanics combined with the fracture evolution law, a specific permeability evolution function is given, realizing the dynamic updating of coal seam permeability during mining disturbance. The coded form is as follows:

[0056]

[0057] In the formula, Real-time permeability of coal, in units of Initial permeability of coal, in units of The result was determined by laboratory seepage experiments; The chemical damage variable is dimensionless and will be solved in subsequent step S2. denoted as the volumetric strain of the coal body, dimensionless, and denoted as the trace of the total strain tensor. The initial porosity of the coal body is dimensionless and determined by laboratory mercury intrusion porosimetry. Temperature coefficient of coal permeability, unit: The volumetric strain of the coal body was calibrated by laboratory variable-temperature seepage experiments. , The trace of the total strain tensor is calculated from the total strain tensor and characterizes the squeezing / propagation effect of the coal body's elastoplastic deformation on the fractures.

[0058] S14: Constructing a constitutive equation for gas adsorption that considers the effect of temperature.

[0059] A constitutive equation for gas adsorption considering the effect of temperature is constructed: by introducing temperature corrections to the Langmuir adsorption parameters, the variation law of gas adsorption / desorption behavior under high geothermal conditions is accurately described.

[0060] Specifically, a temperature-corrected formula for the Langmuir adsorption constant is constructed to reflect the regulatory effect of the temperature field on gas adsorption / desorption. Simultaneously, an engineering calculation form for the chemical potential gradient is provided, achieving coupling between the chemical field, temperature field, and seepage field. The specific coded form is as follows:

[0061]

[0062] In the formula, This refers to the amount of gas adsorption, in units of... ; Original geothermal temperature The Langmuir limiting adsorption capacity at the specified value is expressed in units of 1000 mcg. The adsorption was calibrated by laboratory volumetric adsorption experiments. The activation energy for gas desorption is given by 1000 kilowatts. The result was obtained by fitting the Arrhenius equation to laboratory temperature-switching adsorption experiments. Here is the gas constant, with a value of [value missing]. ; Original geothermal temperature Langmuir adsorption constant at , in units of Heat of adsorption, unit: The result was obtained by fitting the Arrhenius equation to laboratory temperature-varying adsorption experiments.

[0063] (1) and The methods of obtaining these parameters are as follows: all are characteristic parameters of gas adsorption, obtained by fitting the Arrhenius equation through laboratory temperature-varying adsorption experiments.

[0064] (2) Chemical potential gradient Engineering calculation method: In the formula Chemical potential gradient, in units of ; Gas moles, in units of Gas is calculated as methane. ; Standard atmospheric pressure, with a value of [value missing]. ; The enthalpy change of gas adsorption is expressed in units of... The results were determined by laboratory temperature-controlled adsorption experiments. The Hamiltonian operator characterizes the spatial gradient of a physical quantity.

[0065] S2. Based on the chemical potential gradient, a chemical damage evolution model is constructed. By defining chemical damage variables and their deterioration relationship with the mechanical parameters of coal, the degree of damage to the strength of coal by chemical field is quantitatively characterized.

[0066] Step S2 is used to construct a chemical damage evolution model. Specifically, it includes defining chemical damage variables based on the chemical potential gradient. Chemical damage variables This method is used to quantitatively characterize the degree of coal body degradation caused by the initiation and propagation of microcracks induced by coal matrix shrinkage during gas desorption. Its value ranges between no damage and complete destruction. The chemical damage coefficient is calibrated by conducting adsorption / desorption experiments on coal samples under different gas pressures, simultaneously monitoring acoustic emission signals to characterize the degree of microcrack development, fitting the relationship between damage degree and chemical potential gradient, and obtaining the calibrated value of the chemical damage coefficient. A chemical damage variable is established. The dynamic evolution relationship of chemical damage variables is realized by substituting the variation law of gas desorption V with mining disturbance time t into the chemical potential gradient calculation formula. The dynamic updating of chemical damage over mining disturbance time t; the coupling relationship between chemical damage and coal strength is constructed, and the chemical damage variable is introduced into the strength parameters of the coal body to correct the degradation of the original cohesion and original internal friction angle of the coal body, so as to realize the coupling effect of the chemical field on the stress field. The degradation correction is achieved through the combination relationship between the chemical damage variable and the strength degradation coefficient. This solves the engineering pain point that chemical damage variables are difficult to measure in practice, and provides feasible damage parameters for permeability evolution and energy calculation. Specifically, it includes the following steps:

[0067] S21: Define the specific expression for chemical damage variables

[0068] Chemical damage variables The value range represents the degree of coal body deterioration caused by the initiation and propagation of microcracks induced by coal matrix shrinkage during gas desorption. ( This indicates that the coal body has no chemical damage. (This indicates that the coal body has been completely destroyed due to chemical effects), and the specific coding form is:

[0069]

[0070] In the formula, Chemical damage variable, characterizing the degree of deterioration of the coal body caused by the initiation and propagation of microcracks induced by coal matrix shrinkage during gas desorption, with a value range of [value missing]. , ( This indicates that the coal body has no chemical damage. (This indicates that the coal body has been completely destroyed due to chemical effects). Coal chemical damage coefficient, in units of The calibration was performed by a combined laboratory pressure swing adsorption-acoustic emission experiment. : The magnitude of the chemical potential gradient, dimensionless.

[0071] S22: Define the engineering methods for obtaining key parameters

[0072] (1) Chemical damage coefficient Calibration steps: Adsorption / desorption experiments were conducted on coal samples under different gas pressures, and acoustic emission ringing counts (characterizing the degree of microcrack development) were monitored simultaneously. The relationship between the degree of coal damage and the chemical potential gradient was fitted by the acoustic emission ringing counts to obtain... The calibration value is taken as the average value of the coal body in the same mining area as the field application parameter;

[0073] (2) Dynamic evolution of chemical damage variables: by measuring the amount of gas desorption With mining disturbance time The changing pattern (obtained from laboratory desorption experiments) Substituting the curve into the chemical potential gradient calculation formula, we get... Relationship, and thus achieve Dynamic updates. For example... Figure 2 The figure shown is a schematic diagram of the dynamic evolution of chemical damage variables with mining disturbance time in an embodiment of the present invention, in the initial stage (t=0~50h): A slow rise reflects the initial stage of microcrack initiation during gas desorption; the intermediate stage (t=50~200h): Accelerated rise reflects the microcrack propagation stage; later stage (t>200h): The value tends to level off, approaching 1, reflecting that the damage is approaching saturation; where, at point A: t=0, =0 indicates a point where there is no damage. Point B: t=120h. =0.5, indicating more than half of the damage; Point C: t=300h, =0.95 indicates damage saturation.

[0074] in, Mining disturbance time, in units of ; Curve: Curve showing the variation of gas desorption amount with mining disturbance time obtained from laboratory desorption experiments; Relationship: The change of chemical potential gradient with mining disturbance time; The dynamic update relationship of chemical damage variables with mining disturbance time.

[0075] S23: Constructing the coupling relationship between chemical damage and coal strength

[0076] Introducing chemical damage variables The cohesion and internal friction angle of the coal body are modified to mitigate their degradation, thereby achieving coupling between the chemical field and the stress field. Specifically, this is achieved through the following methods:

[0077]

[0078] In the formula, Cohesion of coal after deterioration, in units of ; Original cohesion of coal body, in units of Calibrated by laboratory triaxial compression experiments; The internal friction angle of coal after deterioration, in units of... ; The original internal friction angle of the coal body, in units of... Calibrated by laboratory triaxial compression experiments; Strength degradation coefficient, dimensionless, determined by laboratory damage-strength tests, with a range of values ​​ranging from [value missing]. .

[0079] S3. Based on the field variable distribution obtained by solving the control equations, calculate the total potential energy of the coal-gas system by globally integrating the elastic-plastic deformation energy of the coal body and the internal energy of the gas, and establish an energy mutation instability criterion with the second-order variation of the total potential energy of the system as the core, and determine the critical energy threshold of the outburst hazard.

[0080] Step S3 is used to derive the energy mutation instability criterion based on the global integral of the field variables.

[0081] Based on the energy evolution law of the coal-gas system, and the full-time and space-time field variable distribution (stress field) obtained from step S1, this study focuses on the energy evolution law of the coal-gas system. Gas pressure field Temperature field By calculating the total potential energy of the system through global integration, a stability criterion for the second-order variation of the total potential energy of the system is constructed, and the critical energy threshold for the coal body to change from steady state to unsteady state is determined as a quantitative evaluation index of prominent hazards. The specific integral form of the total potential energy is clarified, laying the foundation for subsequent post-processing analysis of PMNN.

[0082] S31: Determine the specific calculation method for the total potential energy of the system.

[0083] Step S31 calculates the total potential energy of the coal-gas system by globally integrating the elastic-plastic deformation energy of the coal body and the internal energy of the gas. This includes: calculating the potential energy of the coal body by globally integrating the elastic-plastic deformation energy of the coal body. and gas energy The total potential energy of the system consists of two parts; among which, the elastic-plastic deformation energy of the coal body Calculations are based on stress and elastic strain fields, but only the elastic strain portion is considered; the plastic deformation portion is excluded as dissipated energy; gas internal energy. Calculations are performed based on the gas pressure field, temperature field, and gas physical properties, according to the ideal gas law and thermodynamic internal energy formula.

[0084] Specifically: the total potential energy of the coal-gas system Elastic-plastic deformation energy of coal and gas energy It consists of two parts, based on the global integration calculation of field variables, and can be specifically encoded as follows:

[0085]

[0086] In the formula, The total potential energy of the coal-gas system is expressed in units of... It consists of the elastic-plastic deformation energy of the coal body and the internal energy of the gas; The elastic-plastic deformation energy of the coal body is calculated only for the elastic strain portion; the plastic deformation energy is dissipated energy and does not participate in the abrupt change in the system's potential energy. The unit is [unit missing]. ; Internal energy of gas, unit is The derivation is based on the ideal gas law and the thermodynamic internal energy formula; The computational domain for the study area is defined by the coal seam extent determined by on-site geological exploration; The volume is the gas molar volume, in units of Under standard conditions, take ; The total stress tensor of the coal body is one of the field variables for solving the total potential energy of the system. Gas pore pressure is one of the field variables for solving the total potential energy of the system. The real-time temperature of the coal body is one of the field variables for solving the total potential energy of the system. Elastic strain tensor, used to calculate the elastic-plastic deformation energy of coal; Gas molar volume, unit: Under standard conditions, take ; Gas constant, with values ​​ranging from 1 to 10. ; Standard atmosphere, with a value of .

[0087] (2) Calculation principle: The elastic-plastic deformation energy of coal body is calculated only for the elastic strain part, and the plastic deformation energy is the dissipated energy and does not participate in the sudden change of the system potential energy; the internal energy of gas is derived based on the ideal gas law and the thermodynamic internal energy formula.

[0088] Step S32: Establish quantitative criteria for energy mutation instability

[0089] The second variation of the system's total potential energy As a criterion for determining the stability of the coal-gas system, the total potential energy is solved by spatial second-order variational solution. It is the second-order variation of the total potential energy of the system, which serves as the basis for determining the stability of the coal-gas system; Variational operators characterize minute changes in physical quantities. For example... Figure 3 The figure shown is a schematic diagram of the change of the total potential energy U of the system with the mining process in an embodiment of the present invention. The specific determination rule is as follows:

[0090] (1) When At that time, the coal-gas system was in a steady state, microcracks in the coal body developed slowly, and there was no risk of coal seam outburst.

[0091] (2) When At this point, the coal-gas system reaches a critical instability state, and microcracks in the coal rapidly propagate and penetrate. The total potential energy of the system at this time is the critical energy threshold for coal seam outburst danger, denoted as . The unit is That is, the second variation of the total potential energy of the system. The total potential energy of the system at that time;

[0092] (3) When At this time, the coal-gas system is in an unsteady state, the coal body undergoes macroscopic instability and failure, and the coal seam will experience coal and gas outburst.

[0093] S4. Construct and train a physical model-driven neural network. The network uses a high-order time discrete iterative function derived based on the control equations as physical constraints to learn the spatiotemporal mapping law from the field variables and boundary conditions at the current time to the field variables at the next time, and serves as a proxy model for the control equations to achieve rapid prediction of field variables.

[0094] Step S4 is used to train a PMNN adapted to the integer-order THMC system to achieve fast prediction.

[0095] The Physical Model-driven Neural Network (PMNN) proposed in this invention uses the high-order time-discrete iterative laws of integer-order THMC systems as its learning core, integrating the iterative constraints of the physical model with the nonlinear fitting capabilities of the neural network. By embedding the physical laws of the THMC four-field coupling model into the neural network loss function, the model's prediction results not only conform to engineering physics reality but also possess the ability for rapid spatiotemporal mapping. This invention uses this model as a proxy model for the THMC four-field coupling model to replace the step-by-step solution process of traditional numerical simulations, achieving rapid prediction of field variables related to the outburst risk of deep coal seams. Step S4 addresses the problem of low computational efficiency in THMC four-field coupling numerical simulations, which cannot meet the needs of real-time prediction in the field. It adapts the PMNN by weakening fractional-order terminology and clarifying its basis. A high-order time discrete format is used to construct an iterative scheme, which is only used to learn the spatiotemporal iterative laws of integer-order THMC systems. The input / output targets of PMNN are adjusted so that it learns the spatiotemporal mapping from the current field distribution + boundary conditions to the field distribution at the next time step. The total potential energy of the system and the salient hazard criterion are used as independent post-processing analysis modules. The source of training data is optimized to reduce the computational cost and realize the logical closed-loop coupling between PMNN and THMC model. At the same time, the key details of feature selection and numerical format implementation are supplemented in the implementation process.

[0096] Furthermore, the constructed and trained physical model drives the neural network, whose core is positioned as a surrogate model for the four-field coupled control equations of heat, fluid, solidification, and chemical processes. Specifically, this includes: constructing a time iteration scheme for the integer-order heat, fluid, solidification, and chemical system based on a high-order time discrete format, using the spatiotemporal mapping law inherent in this iteration scheme as the learning target of the neural network; the high-order time discrete format adopts... The scheme discretizes the time derivative terms in the governing equations to construct an iterative function from the distribution of field variables and boundary conditions at the previous time step to the distribution of field variables at the next time step. The physical model drives a neural network to learn the input-output mapping relationship of the iterative function, replacing the step-by-step solution process of traditional numerical simulation, and achieving rapid prediction of field variables. Specifically, it includes the following steps:

[0097] S41: Designing the Core Localization and Iteration Scheme for PMNN

[0098] (1) Core positioning: PMNN is a surrogate model for the THMC four-field coupling model, based on A high-order time discrete scheme is used to construct a time iterative scheme for integer-order THMC systems. The spatiotemporal mapping characteristics of this iterative scheme are learned, replacing the step-by-step solution of traditional numerical simulations and improving computational efficiency. This method is not used to solve fractional-order equations; all iterations are based on the integer-order THMC governing equations. It is a high-order time discrete scheme used to construct time iteration schemes for integer-order THMC systems;

[0099] (2) High-order time-discrete iterative scheme: Considering the time evolution characteristics of THMC systems, the following approach is adopted... The format performs high-order discretization on the time term to construct the time step. The following iterative relationship: In the formula For time steps, the unit is ; For the first The distribution of field variables at time t; For the first The distribution of field variables at time t; For computational domain Boundary conditions; Based on The iterative function in the format must be derived based on the time discretization process of the THMC governing equations to ensure the convergence and computational accuracy of the iteration.

[0100] (3) Iteration function The specific derivation process: The derivation needs to be based on the time-discrete process of the THMC governing equations. Taking the gas seepage equation as an example, the derivation will be... Item passed The format is discretized into In the formula for The scheme coefficients, and the other field variables (stress, temperature, damage), are similarly calculated to form a complete... The iterative system is the core work for subsequent numerical implementation; The time derivative term in the gas seepage equation; : No. Gas pore pressure at any given moment; : No. Gas pore pressure at any given moment; The computational domain of the study area is defined by the coal seam range determined by on-site geological exploration.

[0101] S42: Design the PMNN network structure

[0102] A lightweight network structure of "fully connected layer + spatiotemporal mapping layer" is adopted to avoid the curse of dimensionality of high-dimensional field variables, while ensuring the fitting ability of the nonlinear spatiotemporal relationship of the THMC system. The network is an end-to-end structure, with the input being the field variable features and boundary conditions at the current time step, and the output being the field variable features at the next time step. To address the issue of completeness of field variable features, spatial variation features are introduced and a feature selection method is defined.

[0103] Furthermore, the physical model-driven neural network adopts an end-to-end network structure comprising an input layer, hidden layers, a spatiotemporal mapping layer, and an output layer. The input layer receives the field variable features and boundary conditions at the current time step. These field variable features simultaneously include macroscopic statistical features describing the overall level of the field distribution and spatial variation features capturing pre-instability precursors such as damage localization. The hidden layers employ a multi-layer fully connected network for nonlinear feature extraction and mapping. The spatiotemporal mapping layer uses a convolutional structure to maintain the iterative continuity of the field variables over time. The output layer outputs the field variable features for the next time step, with the same dimensionality as the field variable features in the input layer. Figure 4 The diagram shown is a schematic of the PMNN network structure in an embodiment of the present invention. The specific structure is as follows:

[0104] (1) Input layer: The number of nodes is the field variable feature dimension + boundary condition dimension at the current moment; the field variable features take into account "macro-statistical features + spatial variation features". The basic macro-statistical features include the mean / extreme value of stress field, the mean / extreme value of gas pressure field, the mean value of temperature field, and the mean value of chemical damage variable, totaling 8-12; the spatial variation features are the stress gradient, gas pressure gradient, and local extreme value of damage variable that can reflect the precursors of instability, totaling 3-5; the total number of field variable features is controlled within 15; the boundary conditions include ground stress, ground temperature, and mining advance speed, totaling 3-5; in actual development, the core features are screened through feature importance analysis (such as SHAP value, random forest feature weight) and redundant features are eliminated to ensure that key instability signals such as damage localization can be captured;

[0105] (2) Hidden layers: 5 hidden layers are set, with 20 neurons in each layer. The activation function is tanh function to ensure non-linear fitting ability. Dropout layers (dropout=0.1) are added between layers to prevent overfitting.

[0106] (3) Spatiotemporal mapping layer: Set up one one-dimensional convolutional layer with a kernel size of 3 and a stride of 1 to realize the time series mapping of field variables and ensure the temporal continuity of iteration;

[0107] (4) Output layer: The number of nodes is consistent with the feature dimension of the field variables in the input layer. The output is the macroscopic statistical features of the field variables at the next time step + spatial variation features.

[0108] S43: Optimize the training data sources for PMNN

[0109] A hybrid data strategy of "high-fidelity simplified THMC numerical simulation samples + field-measured calibration samples" is adopted to reduce the cost of generating training data while ensuring the physical authenticity and engineering adaptability of the data. Specific data sources are as follows:

[0110] (1) Main training set: Simplified THMC numerical simulation samples, totaling 1000 sets; by simplifying the THMC four-field coupling model through engineering (e.g., ignoring local microscopic effects and using coarse meshing), 800 sets of simplified samples with low computational cost were generated in COMSOL / FLAC3D, and 200 sets of high-fidelity fine simulation samples were generated at the same time. These were mixed and used as the main training set of PMNN. The sample format is " ”;

[0111] (2) Calibration set: field measured samples, a total of 20-50 sets; collect geological exploration data and mining site monitoring data (ground stress, ground temperature, gas pressure, coal body deformation, gas emission) of the target mining area, and obtain the field variable characteristics at the corresponding time through inversion, as the calibration set;

[0112] (3) Data preprocessing: Normalize all samples (to eliminate the influence of dimensions) and remove outliers ( (Guidelines), according to The dataset is divided into training set, validation set, and test set.

[0113] S44: Define the training steps for PMNN

[0114] Step S44 is used to train the physical model-driven neural network, including the following steps: Training data is generated using a hybrid data strategy combining numerical simulation samples and field-measured samples. The main training set is generated from thermal-fluid-solid-chemical numerical simulations, and the calibration set is obtained from field-measured data through inversion. A composite loss function is constructed, comprising field variable fitting loss and physical iteration constraint loss. The physical iteration constraint loss is constructed based on the iteration function and is used to constrain the neural network prediction results to conform to the physical evolution laws of the thermal-fluid-solid-chemical system. An optimization algorithm is used to train the network, and the parameters of the trained model are fine-tuned using the calibration set to control the error between the model prediction results and the field-measured data within a preset range. The trained and calibrated physical model-driven neural network is used for rapid field prediction. The field variable characteristics of its prediction output are used as inputs for subsequent energy mutation instability criterion calculation and hazard classification determination, forming a two-step engineering application mode of rapid prediction and quantitative determination. The specific implementation process is as follows:

[0115] Using the iterative laws of the physical model as constraints, a composite loss function is constructed, and the L-BFGS-B optimizer is used to train the model to ensure that the prediction results of PMNN conform to the physical laws of the THMC system. The specific training steps are as follows:

[0116] Step 1: Model initialization. Randomly initialize the network weights and biases, and set the hyperparameters (learning rate). Maximum number of iterations 1000, batch size );

[0117] Step 2: Construct a composite loss function that incorporates physical iteration constraints and field variable fitting errors to ensure the physical consistency and numerical accuracy of the prediction results. The specific form is as follows:

[0118]

[0119] In the formula, The parameters to be trained in the PMNN model; Physical constraint coefficient, dimensionless, with a value of 0.1; The composite loss function of PMNN consists of field variable fitting loss and physical iteration constraint loss; : Field variable fitting loss, which characterizes the error between the model's predicted field variables and the actual field variables; Physical iteration constraint loss ensures that the model prediction results conform to... Iterative patterns;

[0120] In the formula, Features of field variables predicted by the PMNN model at the next time step; Characteristics of real field variables in numerical simulation; Number of training samples; For sample sequence indexing;

[0121] To ensure that the model prediction results are consistent Iterative law; where, :based on An iterative function in a specific format; The PMNN model predicted the first Time-domain variable distribution; Computational domain Boundary conditions.

[0122] Step 3: Model training. The L-BFGS-B optimizer is used to minimize the composite loss function. Iterative training is performed on the training set until the validation set loss stabilizes (after 50 consecutive iterations of loss variation). Stop training when ( ).

[0123] Step 4: Model calibration. Input the field-measured calibration set into the trained PMNN model, and use Bayesian optimization to fine-tune the model parameters to minimize the error between the model prediction results and the field-measured data. This improves the engineering adaptability of the model;

[0124] Step 5: Model Validation. Input the test set into the calibrated PMNN model and calculate the average relative error of the field variable predictions. If the error... Once the model is trained, it can be used for on-site prediction.

[0125] S45: Develop the prediction and post-processing analysis workflow for PMNN.

[0126] PMNN is only responsible for the rapid spatiotemporal prediction of field variables. The energy mutation criterion and hazard classification are used as independent post-processing analysis modules to realize a two-step engineering application of "rapid prediction - quantitative judgment". The specific process is as follows:

[0127] (1) Input the field variable characteristics (stress, gas pressure, temperature, mean / gradient of chemical damage, local extreme value) and boundary conditions (ground stress, ground temperature, mining advance speed) of the current moment monitored on site into the PMNN model;

[0128] (2) The PMNN model quickly outputs the field variable features of the next time step, and the computational cost of predicting the single-step time is reduced. This improves efficiency by more than 100 times compared to traditional numerical simulation.

[0129] (3) Substitute the field variable characteristics predicted by PMNN into the global integral formula of the total potential energy of the system in step S31 to quickly calculate the total potential energy of the system. ;

[0130] (4) Calculate the second variation of the total potential energy through the post-processing module. , combined The ratio of the values ​​is used to quantitatively determine the outburst risk level of the coal seam based on the grading standards in step S33.

[0131] S5. By comprehensively applying the above steps S1-S4, parameters are calibrated through laboratory experiments, on-site data is collected, critical thresholds are solved through numerical simulation, and a trained physical model is used to drive a neural network for rapid prediction. Based on the comparison between the prediction results and the critical energy threshold, the quantitative classification and engineering early warning of the outburst risk of deep coal seams are finally realized.

[0132] Step S5 is the comprehensive implementation process for quantitative characterization of outburst risk in deep coal seams.

[0133] To improve the reproducibility and engineering feasibility of the method, a simplified standard example is taken with a rectangular coal pillar in a deep coal seam. The entire process, from model building, parameter selection, numerical solution to PMNN training and field prediction, is demonstrated, and critical thresholds are supplemented. An optimized approach that adapts to changes in geological conditions should be developed, and specific steps for on-site engineering applications should be clearly defined to ensure the effective implementation of the method.

[0134] Furthermore, step S5 includes: inputting the current-moment field variable characteristics monitored in real-time into the trained and calibrated physical model to drive the neural network, predicting the field variable distribution at the next moment; selecting the corresponding critical energy threshold according to the geological sub-region to which the predicted location belongs, or estimating the critical energy threshold through a correlation model, and calculating the ratio of the total potential energy of the real-time system to the critical energy threshold; determining the outburst risk level of the coal seam based on the ratio according to a preset quantitative grading standard, and formulating corresponding disaster prevention and control measures accordingly to achieve accurate early warning and prevention of outburst disasters. Specifically, it includes the following steps:

[0135] S51: Conduct simplified standard case demonstrations

[0136] Taking a rectangular coal pillar in a deep coal seam as the research object, the entire process of the method is demonstrated to verify its feasibility and accuracy. Specifically:

[0137] (1) Case Model: Deep Coal Seam Rectangular coal pillar, computational domain Original geostress original geothermal Original gas pressure ;

[0138] (2) Parameter selection: All parameters were calibrated through laboratory experiments, such as , , , ;

[0139] (3) Numerical solution: The THMC four-field coupled model is constructed in COMSOL. The coded control equations of step S1 are used. After meshing, the field variable distribution and critical energy threshold are obtained by solving. J;

[0140] (4) PMNN training: The PMNN model was trained based on 1000 sets of samples generated by numerical simulation. After calibration, the prediction error of the field variables of the model was <6%;

[0141] (5) Prediction and Judgment: Input the field variable characteristics after the coal pillar advances 5 m, PMNN predicts that the gas pressure will rise from 2.5 MPa to 2.8 MPa at the next moment, and the post-processing calculation is as follows. Based on the classification standards, it is determined to be of medium risk and pressure relief and control measures need to be taken.

[0142] S52: Develop specific steps for on-site engineering applications

[0143] Based on the actual conditions of the mining area, feasible on-site engineering application steps were developed, and an optimization approach was added to adapt the critical threshold Ucr to changes in geological conditions. Specifically:

[0144] Step 1: Laboratory experiments, collect coal samples from the target mining area, complete physical and mechanical, gas adsorption, variable temperature seepage, damage-strength and other experiments, and calibrate all parameters of the THMC model and PMNN model;

[0145] Step 2: On-site data collection. Basic data such as geostress, geothermal temperature, gas pressure, and coal seam extent of the study area are obtained through geological exploration and on-site monitoring, and sub-regions are divided according to geological zones (such as by geostress gradient and differences in coal physical properties).

[0146] Step 3: THMC numerical simulation. For each geological sub-region, a THMC four-field coupled model is constructed, and the critical energy threshold of each sub-region is obtained by solving. ,Right now Depending on spatial location Update; simultaneously create Correlation model with core geological parameters: In the formula These are the fitting coefficients. For geostress, For gas pressure, This model represents the elastic modulus of the coal seam and is used for sub-regions where the full simulation was not performed. Quick estimation;

[0147] Step 4: PMNN model training and calibration. The PMNN model is trained based on numerical simulation samples of each sub-region, and the parameters are fine-tuned by combining the field measurement data.

[0148] Step 5: Real-time on-site prediction. Input the field variable characteristics (including spatial variation characteristics) of the current moment monitored on-site into the PMNN model to quickly predict the distribution of field variables at the next moment.

[0149] Step 6: Post-processing judgment. Based on the geological sub-region to which the predicted location belongs, select the corresponding... Or estimate through a correlation model ,calculate And determine the level of prominence based on the classification standards;

[0150] Step 7: Formulate prevention and control measures. Based on the level of hazard, take corresponding depressurization, extraction, and support measures to achieve precise prevention and control of hazardous events.

[0151] Table 1 Comparison of Algorithm Training / Testing Time

[0152]

[0153] To verify the advantages of the algorithm of this invention, a comparative experiment was conducted with conventional algorithms. Table 1 is a comparison table of the training / testing time of the algorithm of this invention and different existing algorithms; Figure 5 This is a comparison of the training loss curves of the algorithm of this invention with those of existing algorithms. From Figure 5 Comparing the loss curves, the Physical Model Driven Neural Network (PMNN) proposed in this invention exhibits the best convergence characteristics. Its loss value decreases steadily and rapidly with the number of iterations, and the final loss value is only 0.03 after 1000 iterations. In contrast, the Feedforward Neural Network (FNN) lacks physical constraints, and its loss fluctuates slightly during training, with the final loss value reaching 0.48. The Long Short-Term Memory Network (LSTM) is affected by both the complexity of sequence modeling and the lack of physical constraints, resulting in more significant loss fluctuations and a final value as high as 0.67. Although the Physical Information Neural Network (PINN) incorporates physical constraints, its convergence speed is slow, and the loss value is still 0.17 after 1000 iterations. From the time performance in Table 1, FNN has the shortest training time (60 seconds) and a single-step testing time of only 50 milliseconds, but its accuracy is significantly inferior. LSTM has a training time of up to 280 seconds and a testing time of 150 milliseconds, and its accuracy is the worst. PINN has the highest training time (360 seconds) and testing time (200 milliseconds), and its accuracy is only slightly better than FNN and LSTM. In contrast, PMNN achieves the best balance between accuracy and efficiency with a training time of 85 seconds and a single-step testing time of 80 milliseconds. It solves the problems of unstable training and large errors of pure data-driven algorithms (FNN, LSTM) and overcomes the defects of slow convergence and long time consumption of PINN. It is more suitable for the engineering field needs of predicting the risk of outbursts in deep coal seams.

[0154] In summary, the quantitative characterization method for outburst risk in deep coal seams based on the multi-field coupling effect of THMC, as presented in this invention, solves the problems of logical discontinuity, low implementability, and low computational efficiency of existing methods by constructing an encodeable set of THMC governing equations, an engineering-feasible chemical damage model, and a PMNN model adapted to integer-order systems. Furthermore, it supplements implementation details such as field variable feature selection, numerical format implementation, and critical threshold adaptability, further enhancing the engineering practicality of the solution. It has significant beneficial effects such as theoretical rigor, engineering implementability, high predictive efficiency, and quantitative results, as detailed below:

[0155] 1. The core positioning of PMNN is clearly defined as a surrogate model for integer-order THMC systems, and a model based on... Formatting iterative functions The derivation approach eliminates the logical gap between the physical model and the AI ​​algorithm; at the same time, the prediction target of PMNN is limited to "macroscopic statistical characteristics of field variables + spatial variation characteristics", and the energy criterion is used as an independent post-processing module, which solves the technical problem that global integral quantities are difficult to predict directly, and realizes the deep and reasonable coupling of multi-field coupling model and AI algorithm.

[0156] 2. The complete coded form of the THMC four-field coupled control equations is given, and all equations have specific mathematical expressions and physical basis; the laboratory calibration methods or engineering inversion methods for all key parameters are clarified, and the core obstacle of obtaining parameters such as chemical potential gradient and chemical damage coefficient is solved; the whole process is demonstrated by simplifying standard examples, which improves the reproducibility of the method, and ordinary engineering technicians can quickly realize the construction and solution of the model based on the examples.

[0157] 3. A hybrid data strategy of "simplified numerical simulation samples + field-measured calibration samples" was adopted to reduce the cost of generating PMNN training data; a lightweight fully connected + spatiotemporal mapping network structure was designed to avoid the curse of dimensionality of high-dimensional field variables; and the computational cost of PMNN single-step prediction was reduced. Compared to traditional THMC numerical simulation, the computational efficiency is improved by more than 100 times, meeting the engineering requirements for real-time on-site prediction. Furthermore, after on-site calibration, the prediction error is significantly reduced. This ensures the accuracy of the prediction.

[0158] 4. Construct an energy mutation criterion based on the second-order variation of the system's total potential energy, using... The ratio of [aspect ratio] to [aspect ratio] is used as a quantitative evaluation index to classify coal seam outburst risk into four distinct levels, resolving the ambiguity problem of traditional qualitative evaluation methods. Simultaneously, the temperature and chemical fields are incorporated into the coupled model, fully considering the disaster-causing mechanisms of the deep "three-high" environment, making the evaluation results more consistent with actual field conditions. This is achieved through "geological zoning solution + geological parameter correlation model." The dynamic updating of spatial location solves the problem of the universality of critical thresholds, providing clear numerical basis for the precise prevention and control of prominent disasters.

[0159] 5. All models in this method can be further developed in commercial numerical simulation software such as COMSOL / FLAC3D without the need for independent development of a simulation platform; the PMNN model has a lightweight structure and can be deployed in the embedded system of the field monitoring terminal to achieve localized rapid prediction; at the same time, the model parameters can be calibrated and adjusted according to the coal body characteristics of different mining areas, and are suitable for the occurrence conditions of different deep coal seams, making it highly practical for engineering applications.

[0160] It should also be noted that, in the embodiments of this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0161] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in the embodiments of this application may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown in this application, but is to be accorded the widest scope consistent with the principles and novel features disclosed in the embodiments of this application.

Claims

1. A quantitative characterization method for outburst risk in deep coal seams based on multi-field coupling effects, characterized in that, Includes the following steps: S1. Establish a set of coded thermal-fluid-solid-chemical four-field coupled control equations. The control equations are based on the occurrence characteristics of deep coal seams. The effective stress equation is modified by introducing temperature stress terms and chemical adsorption stress terms. A permeability dynamic evolution equation containing chemical damage variables is constructed to fully describe the strong coupling mechanism between temperature field, seepage field, stress field and chemical field. S2. Based on the chemical potential gradient, a chemical damage evolution model is constructed. By defining chemical damage variables and their deterioration relationship with the mechanical parameters of coal, the degree of damage to the strength of coal by chemical field is quantitatively characterized. The construction of a chemical damage evolution model specifically includes: Based on the chemical potential gradient, a chemical damage variable is defined. The chemical damage variable is used to quantitatively characterize the degree of deterioration of the coal body caused by the initiation and propagation of microcracks induced by coal matrix shrinkage during gas desorption. Its value range is between no damage and complete destruction. The chemical damage coefficient was calibrated by conducting adsorption / desorption experiments on coal samples under different gas pressures, simultaneously monitoring acoustic emission signals to characterize the degree of microcrack development, fitting the relationship between the degree of damage and the chemical potential gradient, and obtaining the calibrated value of the chemical damage coefficient. A dynamic evolution relationship of chemical damage variables is established by substituting the change law of gas desorption over time into the chemical potential gradient calculation formula, thereby realizing the dynamic updating of chemical damage variables with mining disturbance time. S3. Based on the field variable distribution obtained by solving the control equations, calculate the total potential energy of the coal-gas system by globally integrating the elastic-plastic deformation energy of the coal body and the internal energy of the gas, and establish an energy mutation instability criterion with the second-order variation of the total potential energy of the system as the core, and determine the critical energy threshold of the outburst hazard. S4. Construct and train a physical model-driven neural network. The network uses a high-order time discrete iterative function derived based on the control equations as physical constraints to learn the spatiotemporal mapping law from the field variables and boundary conditions at the current time to the field variables at the next time, and serves as a proxy model for the control equations to achieve rapid prediction of field variables. S5. By comprehensively applying the above steps S1-S4, parameters are calibrated through laboratory experiments, on-site data is collected, critical thresholds are solved through numerical simulation, and a trained physical model is used to drive a neural network for rapid prediction. Based on the comparison between the prediction results and the critical energy threshold, the quantitative classification and engineering early warning of the outburst risk of deep coal seams are finally realized.

2. The method for quantitative characterization of outburst risk in deep coal seams based on multi-field coupling effect according to claim 1, characterized in that, In step S1, a set of coded thermal-fluid-solid-chemical four-field coupled control equations is established, specifically including: The constitutive equation for total coal strain is constructed: the total coal strain consists of four parts: elastic strain, plastic strain, thermal expansion strain, and gas adsorption expansion strain, to describe the deformation behavior of coal under multi-field coupling. A modified effective stress equation was constructed: a temperature stress term and a chemical adsorption stress term were added to the traditional effective stress equation to quantitatively characterize the regulation mechanism of temperature change and gas adsorption / desorption on the effective stress of coal. A dynamic evolution equation for permeability is constructed: by coupling the effects of coal body volumetric strain, chemical damage variables and temperature changes, the dynamic updating of coal seam permeability during mining disturbance is achieved; A constitutive equation for gas adsorption considering the effect of temperature is constructed: by introducing temperature corrections to the Langmuir adsorption parameters, the variation law of gas adsorption / desorption behavior under high geothermal conditions is accurately described.

3. The method for quantitative characterization of outburst risk in deep coal seams based on multi-field coupling effect according to claim 1, characterized in that, Step S2 also includes: A coupling relationship between chemical damage and coal strength is constructed. The chemical damage variable is introduced into the strength parameters of the coal body to modify the original cohesion and original internal friction angle of the coal body, so as to realize the coupling effect of chemical field on stress field. The modification is achieved through the combination relationship between chemical damage variable and strength degradation coefficient.

4. The method for quantitative characterization of outburst risk in deep coal seams based on multi-field coupling effect according to claim 1, characterized in that, In step S3, the calculation of the total potential energy of the coal-gas system by globally integrating the elastic-plastic deformation energy of the coal body and the internal energy of the gas includes: The total potential energy of the system, consisting of the elastic-plastic deformation energy of the coal body and the internal energy of the gas, is calculated by performing global integration over the study area. The elastic-plastic deformation energy of the coal body is calculated based on the stress field and elastic strain field, and only the elastic strain part is included, while the plastic deformation part is not included as dissipated energy. The internal energy of the gas is calculated based on the gas pressure field, temperature field, and gas physical properties, according to the ideal gas law and the thermodynamic internal energy formula.

5. The method for quantitative characterization of outburst risk in deep coal seams based on multi-field coupling effect according to claim 4, characterized in that, In step S3, establishing an energy mutation instability criterion based on the second-order variation of the system's total potential energy and determining the critical energy threshold for protrusion danger specifically includes: The second-order variation of the total potential energy of the system is used as the quantitative basis for determining the stability of the coal-gas system. When the second-order variation is greater than 0, the system is determined to be in a steady state and there is no risk of coal seam outburst. When the second-order variation equals 0, the system is determined to be in a critical instability state, and the total potential energy of the system at this time is determined as the critical energy threshold for coal seam outburst danger. When the second-order variation is less than 0, the system is determined to be in an unsteady state, and a coal and gas outburst will occur in the coal seam.

6. The method for quantitative characterization of outburst risk in deep coal seams based on multi-field coupling effect according to claim 1, characterized in that, In step S4, the constructed and trained physical model drives the neural network, whose core is positioned as a surrogate model for the four-field coupled control equations of heat-fluid-solid-chemical, specifically including: A time iteration scheme for an integer-order heat-fluid-solid-chemical system is constructed based on a high-order time discrete scheme, and the spatiotemporal mapping law contained in the iteration scheme is used as the learning target of the neural network. The higher-order time discretization scheme adopts The scheme discretizes the time derivative terms in the governing equations and constructs an iterative function from the field variable distribution and boundary conditions at the previous time step to the field variable distribution at the next time step. The physical model drives the neural network to learn the input-output mapping relationship of the iterative function, replacing the step-by-step solution process of traditional numerical simulation, and thus achieves rapid prediction of field variables.

7. The method for quantitative characterization of outburst risk in deep coal seams based on multi-field coupling effect according to claim 6, characterized in that, The physical model-driven neural network adopts an end-to-end network structure comprising an input layer, hidden layers, a spatiotemporal mapping layer, and an output layer, wherein: The input layer receives the field variable features and boundary conditions at the current moment. The field variable features include macroscopic statistical features for describing the overall level of the field distribution and spatial variation features for capturing the precursory information of damage localization and instability. The hidden layer consists of multiple fully connected neurons and is used for nonlinear feature extraction and mapping. The spatiotemporal mapping layer adopts a convolutional structure to maintain the iterative continuity of field variables over time. The output layer outputs the field variable features at the next time step, and its dimension is consistent with the field variable feature dimension of the input layer.

8. The method for quantitative characterization of outburst risk in deep coal seams based on multi-field coupling effect according to claim 6, characterized in that, Step S4, which involves building and training the physical model to drive the neural network, also includes the following steps: Training data is generated using a hybrid data strategy that combines numerical simulation samples with field measured samples. The main training set is generated by thermal-fluid-solid-chemical numerical simulation, while the calibration set is obtained by inversion from field measured data. A composite loss function is constructed, which includes field variable fitting loss and physical iteration constraint loss. The physical iteration constraint loss is constructed based on the iteration function and is used to constrain the neural network prediction results to conform to the physical evolution law of the heat-fluid-solid-chemical system. An optimization algorithm is used to train the network, and the parameters of the trained model are fine-tuned using a calibration set to keep the error between the model's prediction results and the actual measured data within a preset range. The trained and calibrated physical model drives the neural network for rapid on-site prediction. The field variable characteristics of its prediction output serve as inputs for subsequent energy mutation instability criterion calculation and hazard classification determination, forming a two-step engineering application mode of rapid prediction and quantitative determination.

9. The method for quantitative characterization of outburst risk in deep coal seams based on multi-field coupling effect according to claim 1, characterized in that, Step S5 includes: The field variable features monitored in real time at the current moment are input into the trained and calibrated physical model to drive the neural network and predict the field variable distribution at the next moment. Based on the geological sub-region to which the predicted location belongs, select the corresponding critical energy threshold, or estimate the critical energy threshold through an association model, and calculate the ratio of the real-time system's total potential energy to the critical energy threshold. Based on the preset quantitative classification standards, the outburst risk level of the coal seam is determined according to the ratio, and corresponding disaster prevention and control measures are formulated accordingly to achieve accurate early warning and prevention and control of outburst disasters.

Images