Mine pressure simulation calibration method based on generative artificial intelligence

Abstract

The invention provides a mine pressure simulation calibration method based on generative artificial intelligence, and relates to the field of coal mine safety. Comprising the following steps: modeling according to mine geological data and performing mine pressure simulation; dynamically acquiring mine pressure observation data and simulation unit parameters in real time; constructing and training a mine pressure simulation real-time calibration model based on the mine pressure observation data and the simulation unit parameters; and performing real-time calibration reasoning on the mine pressure simulation by using the mine pressure simulation real-time calibration model. The problem that the real-time monitoring requirement of mine safety production cannot be met due to obvious time delay between a mine pressure simulation result and actual mine pressure display is solved. The simulation result has small deviation from the actual mine pressure distribution, and the precision is high. And autonomous learning and intelligent adjustment under mine environment change are realized, and complex and changeable mining conditions are easily coped with. And real-time accurate simulation and intelligent calibration of the mine pressure dynamic process are realized, so that more reliable technical support is provided for early warning and prevention of coal mine roof disasters.

Description

Technical Field

[0001] The present invention relates to the field of coal mine safety, and in particular to a mine pressure simulation calibration method based on generative artificial intelligence. Background Art

[0002] In the field of coal mine safety, mine pressure simulation and calibration technology is a key technical means to ensure mine stability and safe production. Mine pressure refers to the pressure generated in a mine due to the combined action of rock gravity and tectonic stress. Its dynamic changes directly affect the stability of the mine roof, the control effect of the tunnel surrounding rock, and overall safe production. With the continuous increase in the depth and intensity of coal mining, the characteristics of mine pressure are becoming increasingly complex, and higher requirements are placed on the accuracy, real-time and adaptability of mine pressure simulation technology. Traditional mine pressure simulation methods mainly include: simulation methods based on traditional physical models, which simulate the distribution and evolution of mine pressure by establishing rock mechanics constitutive equations and boundary conditions, combined with numerical calculation methods.

[0003] However, these methods have problems such as complex model construction and low computational efficiency, and it is difficult to accurately reflect the heterogeneity of mine geological conditions, the anisotropy of rock mass structure, and the dynamic change characteristics during the mining process. Based on traditional machine learning methods, machine learning algorithms such as linear regression and decision trees were used in the early days to predict mine pressure. Although such methods have reduced the modeling complexity to a certain extent, they still rely on a large amount of manual feature engineering and have obvious limitations in dealing with the complex nonlinear relationship between mine pressure and multiple factors such as geological conditions and mining technology; methods based on traditional data analysis mainly use statistical analysis of historical monitoring data to infer the law of mine pressure changes. Such methods lack real-time dynamic adjustment capabilities and are difficult to adapt to rapid changes in the mine environment. Especially when mining conditions change suddenly or geological conditions are complex, their prediction accuracy and reliability are significantly reduced.

[0004] Comprehensive analysis shows that the traditional mine pressure simulation method has the following major technical defects:

[0005] Insufficient real-time performance: It is difficult to achieve real-time collection, processing and analysis of mine pressure observation data, resulting in a significant time lag between simulation results and actual mine pressure, which cannot meet the real-time monitoring needs of mine safety production.

[0006] Limited accuracy: When dealing with the coupling of multiple factors such as rock mass heterogeneity, mining disturbance, and hydrogeological conditions, it is difficult to accurately characterize the dynamic evolution of mine pressure, resulting in a large deviation between the simulation results and the actual mine pressure distribution.

[0007] Poor adaptability: A large amount of manual intervention is required for model adjustment and parameter optimization. It is unable to achieve autonomous learning and intelligent adjustment under changing mine environment, and it is difficult to cope with complex and changing mining conditions. Summary of the Invention

[0008] In view of the above problems, the present invention is proposed to provide a mine pressure simulation calibration method based on generative artificial intelligence that solves the above problems or partially solves the above problems.

[0009] A first aspect of an embodiment of the present invention provides a mine pressure simulation calibration method based on generative artificial intelligence, the mine pressure simulation calibration method comprising:

[0010] Modeling and conducting mine pressure simulation based on mine geological data;

[0011] Real-time dynamic acquisition of mine pressure observation data and simulation unit parameters;

[0012] Based on the mine pressure observation data and the simulation unit parameters, a mine pressure simulation real-time calibration model is constructed and trained;

[0013] The mine pressure simulation real-time calibration model is used to perform real-time calibration reasoning on the mine pressure simulation.

[0014] Optionally, modeling and performing mine pressure simulation based on mine geological data includes:

[0015] Using at least one of electrical, magnetic, gravity, seismic, and radar methods, the detection obtains location information for the mine geological structure modeling, the location information including location information corresponding to coal and rock layers, faults, and collapse columns;

[0016] Combining the mine geological structure modeling location information with the mine geological data modeling to obtain a mine pressure model;

[0017] Performing mine pressure simulation on the mine pressure model using mine pressure simulation software;

[0018] The mine geological data includes laboratory test data of coal seam core samples on density, water content, sulfur content, ash content, elastic modulus, Poisson's ratio, compressive strength, tensile strength, shear modulus, and bulk modulus;

[0019] The mine pressure simulation software includes: COMSOL, ABAQUS, ANSYS, PFC or FLAC.

[0020] Optionally, real-time dynamic acquisition of mine pressure observation data and simulation unit parameters includes:

[0021] The mine pressure observation data is obtained dynamically and in real time through various sensors or databases;

[0022] Dynamically acquiring the simulation unit parameters from the database in real time;

[0023] The mine pressure observation data include: working face support resistance, tunnel surrounding rock mine pressure, microseismic data, roof settlement, roof stress, mining progress, and drill cuttings volume;

[0024] The simulation unit parameters include elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength and density.

[0025] Optionally, based on the mine pressure observation data and the simulation unit parameters, constructing and training a mine pressure simulation real-time calibration model includes:

[0026] Construct a physical observation encoding network, a physical observation decoding network, and a conditional generation network;

[0027] Based on the mine pressure observation data and the simulation unit parameters, the physical observation encoding network, the physical observation decoding network and the condition generation network are trained to obtain the mine pressure simulation real-time calibration model.

[0028] Optionally, training a physical observation encoding network, a physical observation decoding network, and a conditional generation network includes:

[0029] Using a fully connected neural network, a convolutional neural network, a recurrent neural network, or an attention mechanism neural network, the physical observation encoding network and the physical observation decoding network are trained;

[0030] The conditional generative network is trained using a conditional generative adversarial network, a conditional variational autoencoder, or a conditional diffusion model.

[0031] Optionally, training a physical observation encoding network and a physical observation decoding network includes:

[0032] The mine pressure observation data is used as input of the physical observation coding network, and after being processed by the physical observation coding network, a hidden representation of the mine pressure observation data is output;

[0033] Using the hidden representation of the mine pressure observation data as input to the physical observation decoding network, and outputting the mine pressure prediction data after being processed by the physical observation decoding network;

[0034] Optimizing the self-representation loss function of the mine pressure observation data by a back-propagation algorithm so that the output results of the physical observation encoding network and the physical observation decoding network are accurate;

[0035] Among them, the self-representation loss function Loss s for:

[0036] Loss s =Decoder(Encoder(k))-k

[0037] In the above formula, k represents the mine pressure observation data, Encoder represents the physical observation encoding network, and Decoder represents the physical observation decoding network.

[0038] Optionally, training a conditional generative network includes:

[0039] Taking the standard normal distribution data and the hidden representation of the mine pressure observation data as input, after being processed by the conditional generation network, the simulation prediction parameters are output;

[0040] Based on the simulation prediction parameters and the simulation unit parameters, the generation loss function of the simulation prediction parameters and the simulation unit parameters is optimized by a back propagation algorithm so that the output result of the conditional generation network is accurate.

[0041] Optionally, the standard normal distribution data and the hidden representation of the mine pressure observation data are used as input, and after being processed by the conditional generative network, the simulation prediction parameters are output, including:

[0042] Generate n standard normal distribution samples;

[0043] Inputting the n standard normal distribution samples into the physical observation encoder respectively to obtain n hidden representations;

[0044] Input each standard normal distribution sample and each hidden representation into the conditional generative network to obtain n initial prediction parameters;

[0045] Calculate the average value of the n initial prediction parameters and use the average value as the simulation prediction parameter.

[0046] Optionally, the generated loss function Loss g for:

[0047]

[0048] In the above formula, z i Satisfies the standard normal distribution, i=1,2,…,n, Gen represents the conditional generation network, and x represents the simulation unit parameter.

[0049] Optionally, the loss function of the mine pressure simulation real-time calibration model obtained by training is:

[0050]

[0051] In the above formula, Loss represents the loss function of the mine pressure simulation real-time calibration model.

[0052] A second aspect of an embodiment of the present invention provides a mine pressure simulation and calibration system based on generative artificial intelligence, the mine pressure simulation and calibration system comprising:

[0053] Modeling and simulation module, used to model and simulate mine pressure based on mine geological data;

[0054] Acquisition module, used to dynamically acquire mine pressure observation data and simulation unit parameters in real time;

[0055] Constructing a training module for constructing and training a mine pressure simulation real-time calibration model based on the mine pressure observation data and the simulation unit parameters;

[0056] The calibration reasoning module is used to perform real-time calibration reasoning on the mine pressure simulation using the mine pressure simulation real-time calibration model.

[0057] Optionally, the modeling and simulation module is specifically used to:

[0058] Using at least one of electrical, magnetic, gravity, seismic, and radar methods, the detection obtains location information for the mine geological structure modeling, the location information including location information corresponding to coal and rock layers, faults, and collapse columns;

[0059] Combining the mine geological structure modeling location information with the mine geological data modeling to obtain a mine pressure model;

[0060] Performing mine pressure simulation on the mine pressure model using mine pressure simulation software;

[0061] The mine geological data includes laboratory test data of coal seam core samples on density, water content, sulfur content, ash content, elastic modulus, Poisson's ratio, compressive strength, tensile strength, shear modulus, and bulk modulus;

[0062] The mine pressure simulation software includes: COMSOL, ABAQUS, ANSYS, PFC or FLAC.

[0063] Optionally, the acquisition module is specifically used to:

[0064] The mine pressure observation data is obtained dynamically and in real time through various sensors or databases;

[0065] Dynamically acquiring the simulation unit parameters from the database in real time;

[0066] The mine pressure observation data include: working face support resistance, tunnel surrounding rock mine pressure, microseismic data, roof settlement, roof stress, mining progress, and drill cuttings volume;

[0067] The simulation unit parameters include elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength and density.

[0068] Optionally, building a training module includes:

[0069] Construction submodules for constructing physical observation encoding network, physical observation decoding network and conditional generation network;

[0070] The training submodule is used to train the physical observation encoding network, the physical observation decoding network and the condition generation network based on the mine pressure observation data and the simulation unit parameters to obtain the mine pressure simulation real-time calibration model.

[0071] Optionally, the training submodule is specifically used to:

[0072] Using a fully connected neural network, a convolutional neural network, a recurrent neural network, or an attention mechanism neural network, the physical observation encoding network and the physical observation decoding network are trained;

[0073] The conditional generative network is trained using a conditional generative adversarial network, a conditional variational autoencoder, or a conditional diffusion model.

[0074] Optionally, the training submodule is further configured to:

[0075] The mine pressure observation data is used as input of the physical observation coding network, and after being processed by the physical observation coding network, a hidden representation of the mine pressure observation data is output;

[0076] Using the hidden representation of the mine pressure observation data as input to the physical observation decoding network, and outputting the mine pressure prediction data after being processed by the physical observation decoding network;

[0077] Optimizing the self-representation loss function of the mine pressure observation data by a back-propagation algorithm so that the output results of the physical observation encoding network and the physical observation decoding network are accurate;

[0078] Among them, the self-representation loss function Loss s for:

[0079] Loss s =Decoder(Encoder(k))-k

[0080] In the above formula, k represents the mine pressure observation data, Encoder represents the physical observation encoding network, and Decoder represents the physical observation decoding network.

[0081] Optionally, the training submodule is further configured to:

[0082] Taking the standard normal distribution data and the hidden representation of the mine pressure observation data as input, after being processed by the conditional generation network, the simulation prediction parameters are output;

[0083] Based on the simulation prediction parameters and the simulation unit parameters, optimizing the generation loss function of the simulation prediction parameters and the simulation unit parameters through a back propagation algorithm so that the output result of the conditional generation network is accurate;

[0084] The standard normal distribution data and the hidden representation of the mine pressure observation data are used as input, and after being processed by the conditional generation network, the simulation prediction parameters are output, including:

[0085] Generate n standard normal distribution samples;

[0086] Inputting the n standard normal distribution samples into the physical observation encoder respectively to obtain n hidden representations;

[0087] Input each standard normal distribution sample and each hidden representation into the conditional generative network to obtain n initial prediction parameters;

[0088] Calculating an average value of the n initial prediction parameters and using the average value as the simulation prediction parameter;

[0089] The generated loss function Loss g for:

[0090]

[0091] In the above formula, z i Satisfying the standard normal distribution, i=1,2,…,n, Gen represents the conditional generation network, and x represents the simulation unit parameter;

[0092] The loss function of the mine pressure simulation real-time calibration model obtained by training is:

[0093]

[0094] In the above formula, Loss represents the loss function of the mine pressure simulation real-time calibration model.

[0095] The mine pressure simulation calibration method based on generative artificial intelligence provided by the present invention includes: modeling and performing mine pressure simulation according to mine geological data; dynamically acquiring mine pressure observation data and simulation unit parameters in real time; constructing and training a mine pressure simulation real-time calibration model based on the mine pressure observation data and simulation unit parameters; and performing real-time calibration reasoning on the mine pressure simulation using the mine pressure simulation real-time calibration model.

[0096] The mine pressure simulation and calibration method based on generative artificial intelligence proposed in the present invention creatively draws on generative artificial intelligence technology, utilizing its significant advantages in complex data relationship modeling and high-quality content generation. By combining generative artificial intelligence technology with mine pressure simulation, real-time collection, processing and analysis of mine pressure observation data are realized, and the mine pressure simulation calibration model is used to calibrate and infer the mine pressure simulation, thereby solving the problem that there is a significant time lag between the mine pressure simulation results and the actual mine pressure manifestation, and the real-time monitoring needs of mine safety production cannot be met.

[0097] Because it accurately depicts the dynamic evolution of mine pressure, the simulation results have a small deviation from the actual mine pressure distribution and high accuracy. It also eliminates the need for extensive manual intervention for model adjustment and parameter optimization, enabling autonomous learning and intelligent adjustment under changing mine environments. This allows for easy handling of complex and changing mining conditions, enabling real-time, accurate simulation and intelligent calibration of the dynamic mine pressure process. This provides more reliable technical support for coal mine roof disaster warning and prevention, and is highly practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0099] Figure 1 This is a flow chart of a mine pressure simulation calibration method based on generative artificial intelligence according to an embodiment of the present invention;

[0100] Figure 2 This is a block diagram of a mine pressure simulation and calibration system based on generative artificial intelligence in an embodiment of the present invention. DETAILED DESCRIPTION

[0101] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present invention, are only part of the embodiments of the present invention, not all of the embodiments, and are not intended to limit the present invention.

[0102] The present invention proposes a mine pressure simulation calibration method based on generative artificial intelligence, referring to Figure 1 The flowchart of the mine pressure simulation calibration method shown in FIG. 1 includes:

[0103] Step 101: Modeling and performing mine pressure simulation based on mine geological data.

[0104] Since the present invention calibrates the mine pressure simulation, it is first necessary to construct a mine pressure model and then perform mine pressure simulation based on the relevant data. The subsequent mine pressure simulation real-time calibration model performs real-time calibration and reasoning on the mine pressure simulation results of the mine pressure model constructed in step 101.

[0105] A better method for modeling and simulating mine pressure based on mine geological data includes:

[0106] Using at least one of electrical, magnetic, gravity, seismic, and radar methods, or a combination of multiple methods, the location information for mine geological structure modeling is obtained. This location information includes the location information corresponding to coal and rock layers, faults, and collapse columns. Whether using a single detection method or a combination of multiple detection methods, the goal is to ensure the accuracy of the location information for mine geological structure modeling, thereby ensuring the accuracy of subsequent mine pressure modeling and simulation.

[0107] After obtaining the mine geological structure modeling location information, the mine geological structure modeling location information and the mine geological data are combined to perform modeling to obtain a mine pressure model; and then the mine pressure simulation software is used to perform mine pressure simulation on the mine pressure model.

[0108] Among them, mine geological data include laboratory test density, water content, sulfur content, ash content, as well as elastic modulus, Poisson's ratio, compressive strength, tensile strength, shear modulus and bulk modulus of coal seam core samples.

[0109] Mine pressure simulation software includes: COMSOL, ABAQUS, ANSYS, PFC or FLAC, etc.

[0110] Step 102: dynamically acquiring mine pressure observation data and simulation unit parameters in real time.

[0111] While performing mine pressure modeling and simulation, mine pressure observation data and simulation unit parameters are acquired in real time and dynamically. A better way to acquire these data in real time and dynamically includes:

[0112] Through various sensors or databases, the mine pressure observation data is obtained in real time and dynamically; and the simulation unit parameters are obtained in real time and dynamically from the database.

[0113] Among them, the mine pressure observation data include: working face support resistance, tunnel surrounding rock mine pressure, microseismic data, roof settlement, roof stress, mining progress, and drill cuttings volume; the simulation unit parameters include: elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength and density.

[0114] Step 103: Based on the mine pressure observation data and simulation unit parameters, a mine pressure simulation real-time calibration model is constructed and trained.

[0115] After obtaining the mine pressure observation data and simulation unit parameters, a mine pressure simulation real-time calibration model can be constructed and trained based on the mine pressure observation data and simulation unit parameters. A better method for constructing and training the mine pressure simulation real-time calibration model includes:

[0116] First, a physical observation encoding network, a physical observation decoding network, and a conditional generation network are constructed. Then, based on the mine pressure observation data and simulation unit parameters, these networks are trained to obtain a real-time calibration model for mine pressure simulation. In other words, the real-time calibration model for mine pressure simulation in this embodiment of the present invention is obtained by training three networks: the physical observation encoding network, the physical observation decoding network, and the conditional generation network.

[0117] For the above three networks, the better ones can use fully connected neural networks (BPNN), convolutional neural networks (CNN), recurrent neural networks (RNN) or attention mechanism neural networks (Transformer) to train physical observation encoding networks and physical observation decoding networks; and use conditional generative adversarial networks (CGAN), conditional variational autoencoders (CVAE) or conditional diffusion models (CDMs) to train conditional generative networks.

[0118] Specifically, the better methods for training the physical observation encoding network and the physical observation decoding network include:

[0119] The physical observation encoding network uses the observed mine pressure data as input, which is processed by the network to output a hidden representation of the observed mine pressure data. The physical observation decoding network also uses the hidden representation of the observed mine pressure data as input, which is processed by the network to output predicted mine pressure data. During the training process, a backpropagation algorithm is also used to optimize the self-representation loss function of the observed mine pressure data. This self-representation loss function can make the output results of the physical observation encoding network and the physical observation decoding network more accurate.

[0120] Among them, the self-representation loss function Loss used to optimize the mine pressure observation data s for:

[0121] Loss s =Decoder(Encoder(k))-k

[0122] In the above formula, k represents the mine pressure observation data, Encoder represents the physical observation encoding network, and Decoder represents the physical observation decoding network.

[0123] For training conditional generative networks, a better approach includes:

[0124] The hidden representation of standard normal distribution data and mine pressure observation data is used as input, and after processing by the conditional generative network, the simulation prediction parameters are output; then, based on the simulation prediction parameters and the obtained simulation unit parameters, the generation loss function of the simulation prediction parameters and the simulation unit parameters is optimized through the back propagation algorithm. This generation loss function can make the output results of the conditional generative network more accurate.

[0125] The specific generation method for the simulation prediction parameters output by the conditional generation network includes:

[0126] First, generate n standard normal distribution samples; then input the n standard normal distribution samples into the aforementioned physical observation encoder respectively, so as to obtain the hidden representation of the n standard normal distribution samples; then input each standard normal distribution sample and the hidden representation of each standard normal distribution sample into the conditional generation network respectively, and n initial prediction parameters can be obtained; finally, calculate the average value of the n initial prediction parameters, and use the average value as the final simulation prediction parameter.

[0127] The loss function Loss generated by optimizing the simulation prediction parameters and simulation unit parameters used in the above training process g for:

[0128]

[0129] In the above formula, z i Satisfies the standard normal distribution, i=1,2,…,n, Gen represents the conditional generation network, and x represents the simulation unit parameter.

[0130] The self-representation loss function Loss is obtained by the above method s And generate loss function Loss g Afterwards, in order to ensure the accuracy of the mine pressure simulation real-time calibration model, a loss function used in training the mine pressure simulation real-time calibration model is also required, which is:

[0131]

[0132] In the above formula, Loss represents the loss function of the mine pressure simulation real-time calibration model.

[0133] Through the above explanations and illustrations, it can be known that the real-time calibration model for mine pressure simulation proposed in the present invention draws on generative artificial intelligence technology and utilizes its significant advantages in complex data relationship modeling and high-quality content generation. For example, large language models represented by ChatGPT have achieved remarkable results in natural language processing, image generation and other fields through deep learning frameworks and massive data training. Generative artificial intelligence technology can process high-dimensional, nonlinear, and unstructured data, and has powerful feature extraction and pattern recognition capabilities. This provides a new idea for technological innovation in the field of coal mine safety: by combining generative artificial intelligence technology with mine pressure simulation, real-time collection, processing and analysis of mine pressure observation data can be achieved, and the mine pressure simulation calibration model is used to calibrate and infer the mine pressure simulation, solving the problem of significant time lag between the mine pressure simulation results and the actual mine pressure manifestation, which cannot meet the real-time monitoring needs of mine safety production. In addition, because the mine pressure simulation calibration model accurately depicts the dynamic evolution law of mine pressure, the deviation between the simulation results and the actual mine pressure distribution is small and the accuracy is high. At the same time, it does not require a large amount of manual intervention for model adjustment and parameter optimization, and can achieve autonomous learning and intelligent adjustment under changes in the mine environment, easily cope with complex and changeable mining conditions, and achieve real-time and accurate simulation and intelligent calibration of the dynamic process of mine pressure, thereby providing more reliable technical support for coal mine roof disaster warning and prevention.

[0134] Step 104: using the mine pressure simulation real-time calibration model to perform real-time calibration reasoning on the mine pressure simulation.

[0135] After the mine pressure simulation real-time calibration model is trained, the mine pressure simulation real-time calibration model can be used to perform real-time calibration reasoning on the mine pressure simulation, thereby obtaining a mine pressure simulation result with better accuracy and higher precision.

[0136] Based on the above-mentioned mine pressure simulation calibration method, the embodiment of the present invention also proposes a mine pressure simulation calibration system based on generative artificial intelligence, referring to Figure 2 The block diagram of the mine pressure simulation calibration system shown in FIG. 1 includes:

[0137] Modeling and simulation module 210, used to model and perform mine pressure simulation based on mine geological data;

[0138] Acquisition module 220, for real-time dynamic acquisition of mine pressure observation data and simulation unit parameters;

[0139] A training module 230 is constructed to construct and train a real-time calibration model for mine pressure simulation based on the mine pressure observation data and the simulation unit parameters;

[0140] The calibration reasoning module 240 is configured to perform real-time calibration reasoning on the mine pressure simulation using the mine pressure simulation real-time calibration model.

[0141] Optionally, the modeling and simulation module 210 is specifically configured to:

[0142] Using at least one of electrical, magnetic, gravity, seismic, and radar methods, the detection obtains location information for the mine geological structure modeling, the location information including location information corresponding to coal and rock layers, faults, and collapse columns;

[0143] Combining the mine geological structure modeling location information with the mine geological data modeling to obtain a mine pressure model;

[0144] Performing mine pressure simulation on the mine pressure model using mine pressure simulation software;

[0145] The mine geological data includes laboratory test data of coal seam core samples on density, water content, sulfur content, ash content, elastic modulus, Poisson's ratio, compressive strength, tensile strength, shear modulus, and bulk modulus;

[0146] The mine pressure simulation software includes: COMSOL, ABAQUS, ANSYS, PFC or FLAC.

[0147] Optionally, the acquisition module 220 is specifically configured to:

[0148] The mine pressure observation data is obtained dynamically and in real time through various sensors or databases;

[0149] Dynamically acquiring the simulation unit parameters from the database in real time;

[0150] The mine pressure observation data include: working face support resistance, tunnel surrounding rock mine pressure, microseismic data, roof settlement, roof stress, mining progress, and drill cuttings volume;

[0151] The simulation unit parameters include: elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength and density.

[0152] Optionally, the constructing training module 230 includes:

[0153] Construction submodules for constructing physical observation encoding network, physical observation decoding network and conditional generation network;

[0154] The training submodule is used to train the physical observation encoding network, the physical observation decoding network and the condition generation network based on the mine pressure observation data and the simulation unit parameters to obtain the mine pressure simulation real-time calibration model.

[0155] Optionally, the training submodule is specifically used to:

[0156] Using a fully connected neural network, a convolutional neural network, a recurrent neural network, or an attention mechanism neural network, the physical observation encoding network and the physical observation decoding network are trained;

[0157] The conditional generative network is trained using a conditional generative adversarial network, a conditional variational autoencoder, or a conditional diffusion model.

[0158] Optionally, the training submodule is further configured to:

[0159] The mine pressure observation data is used as input of the physical observation coding network, and after being processed by the physical observation coding network, a hidden representation of the mine pressure observation data is output;

[0160] Using the hidden representation of the mine pressure observation data as input to the physical observation decoding network, and outputting the mine pressure prediction data after being processed by the physical observation decoding network;

[0161] Optimizing the self-representation loss function of the mine pressure observation data by a back-propagation algorithm so that the output results of the physical observation encoding network and the physical observation decoding network are accurate;

[0162] Among them, the self-representation loss function Loss s for:

[0163] Loss s =Decoder(Encoder(k))-k

[0164] In the above formula, k represents the mine pressure observation data, Encoder represents the physical observation encoding network, and Decoder represents the physical observation decoding network.

[0165] Optionally, the training submodule is further configured to:

[0166] Taking the standard normal distribution data and the hidden representation of the mine pressure observation data as input, after being processed by the conditional generation network, the simulation prediction parameters are output;

[0167] Based on the simulation prediction parameters and the simulation unit parameters, optimizing the generation loss function of the simulation prediction parameters and the simulation unit parameters through a back propagation algorithm so that the output result of the conditional generation network is accurate;

[0168] The standard normal distribution data and the hidden representation of the mine pressure observation data are used as input, and after being processed by the conditional generation network, the simulation prediction parameters are output, including:

[0169] Generate n standard normal distribution samples;

[0170] Inputting the n standard normal distribution samples into the physical observation encoder respectively to obtain n hidden representations;

[0171] Input each standard normal distribution sample and each hidden representation into the conditional generative network to obtain n initial prediction parameters;

[0172] Calculating an average value of the n initial prediction parameters and using the average value as the simulation prediction parameter;

[0173] The generated loss function Loss g for:

[0174]

[0175] In the above formula, z i Satisfying the standard normal distribution, i=1,2,…,n, Gen represents the conditional generation network, and x represents the simulation unit parameter;

[0176] The loss function of the mine pressure simulation real-time calibration model obtained by training is:

[0177]

[0178] In the above formula, Loss represents the loss function of the mine pressure simulation real-time calibration model.

[0179] Through the above embodiments, the mine pressure simulation calibration method based on generative artificial intelligence provided by the present invention includes: modeling and performing mine pressure simulation according to mine geological data; dynamically acquiring mine pressure observation data and simulation unit parameters in real time; constructing and training a mine pressure simulation real-time calibration model based on the mine pressure observation data and simulation unit parameters; and performing real-time calibration inference on the mine pressure simulation using the mine pressure simulation real-time calibration model.

[0180] The mine pressure simulation and calibration method based on generative artificial intelligence proposed in the present invention creatively draws on generative artificial intelligence technology, utilizing its significant advantages in complex data relationship modeling and high-quality content generation. By combining generative artificial intelligence technology with mine pressure simulation, real-time collection, processing and analysis of mine pressure observation data are realized, and the mine pressure simulation calibration model is used to calibrate and infer the mine pressure simulation, thereby solving the problem that there is a significant time lag between the mine pressure simulation results and the actual mine pressure manifestation, and the real-time monitoring needs of mine safety production cannot be met.

[0181] Because it accurately depicts the dynamic evolution of mine pressure, the simulation results have a small deviation from the actual mine pressure distribution and high accuracy. It also eliminates the need for extensive manual intervention for model adjustment and parameter optimization, enabling autonomous learning and intelligent adjustment under changing mine environments. It easily copes with complex and changing mining conditions, achieving real-time, accurate simulation and intelligent calibration of the dynamic process of mine pressure, thereby providing more reliable technical support for coal mine roof disaster warning and prevention, and possessing high practicality.

[0182] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0183] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A mine pressure simulation calibration method based on generative artificial intelligence, characterized in that: The mine pressure simulation calibration method includes: Modeling and conducting mine pressure simulation based on mine geological data; Real-time dynamic acquisition of mine pressure observation data and simulation unit parameters; Based on the mine pressure observation data and the simulation unit parameters, a mine pressure simulation real-time calibration model is constructed and trained; The mine pressure simulation real-time calibration model is used to perform real-time calibration reasoning on the mine pressure simulation.

2. The mine pressure simulation calibration method according to claim 1, characterized in that: Modeling and mine pressure simulation based on mine geological data, including: Using at least one of electrical, magnetic, gravity, seismic, and radar methods, the detection obtains location information for the mine geological structure modeling, the location information including location information corresponding to coal and rock layers, faults, and collapse columns; Combining the mine geological structure modeling location information with the mine geological data modeling to obtain a mine pressure model; Performing mine pressure simulation on the mine pressure model using mine pressure simulation software; The mine geological data includes laboratory test data of coal seam core samples on density, water content, sulfur content, ash content, elastic modulus, Poisson's ratio, compressive strength, tensile strength, shear modulus, and bulk modulus; The mine pressure simulation software includes: COMSOL, ABAQUS, ANSYS, PFC or FLAC.

3. The mine pressure simulation calibration method according to claim 1, characterized in that: Real-time dynamic acquisition of mine pressure observation data and simulation unit parameters, including: The mine pressure observation data is obtained dynamically and in real time through various sensors or databases; Dynamically acquiring the simulation unit parameters from the database in real time; The mine pressure observation data include: working face support resistance, tunnel surrounding rock mine pressure, microseismic data, roof settlement, roof stress, mining progress, and drill cuttings volume; The simulation unit parameters include elastic modulus, Poisson's ratio, cohesion, internal friction angle, tensile strength and density.

4. The mine pressure simulation calibration method according to claim 1, characterized in that: Based on the mine pressure observation data and the simulation unit parameters, a mine pressure simulation real-time calibration model is constructed and trained, including: Construct a physical observation encoding network, a physical observation decoding network, and a conditional generation network; Based on the mine pressure observation data and the simulation unit parameters, the physical observation encoding network, the physical observation decoding network and the condition generation network are trained to obtain the mine pressure simulation real-time calibration model.

5. The mine pressure simulation calibration method according to claim 4, characterized in that: Training the physical observation encoding network, physical observation decoding network, and conditional generation network, including: Using a fully connected neural network, a convolutional neural network, a recurrent neural network, or an attention mechanism neural network, the physical observation encoding network and the physical observation decoding network are trained; The conditional generative network is trained using a conditional generative adversarial network, a conditional variational autoencoder, or a conditional diffusion model.

6. The mine pressure simulation calibration method according to claim 4, characterized in that: Training the physical observation encoding network and physical observation decoding network, including: The mine pressure observation data is used as input of the physical observation coding network, and after being processed by the physical observation coding network, a hidden representation of the mine pressure observation data is output; Using the hidden representation of the mine pressure observation data as input to the physical observation decoding network, and outputting the mine pressure prediction data after being processed by the physical observation decoding network; Optimizing the self-representation loss function of the mine pressure observation data by a back-propagation algorithm so that the output results of the physical observation encoding network and the physical observation decoding network are accurate; Among them, the self-representation loss function Loss s for: Loss s =Decoder(Encoder(k))-k In the above formula, k represents the mine pressure observation data, Encoder represents the physical observation encoding network, and Decoder represents the physical observation decoding network.

7. The mine pressure simulation calibration method according to claim 6, characterized in that: Training a conditional generative network, including: Taking the standard normal distribution data and the hidden representation of the mine pressure observation data as input, after being processed by the conditional generation network, the simulation prediction parameters are output; Based on the simulation prediction parameters and the simulation unit parameters, the generation loss function of the simulation prediction parameters and the simulation unit parameters is optimized by a back propagation algorithm so that the output result of the conditional generation network is accurate.

8. The mine pressure simulation calibration method according to claim 7, characterized in that: The standard normal distribution data and the hidden representation of the mine pressure observation data are used as input, and after being processed by the conditional generation network, the simulation prediction parameters are output, including: Generate n standard normal distribution samples; Inputting the n standard normal distribution samples into the physical observation encoder respectively to obtain n hidden representations; Input each standard normal distribution sample and each hidden representation into the conditional generative network to obtain n initial prediction parameters; Calculate the average value of the n initial prediction parameters and use the average value as the simulation prediction parameter.

9. The mine pressure simulation calibration method according to claim 7, characterized in that: The generated loss function Loss g for: In the above formula, z i Satisfies the standard normal distribution, i=1,2,…,n, Gen represents the conditional generation network, and x represents the simulation unit parameter.

10. The mine pressure simulation calibration method according to claim 9, characterized in that: The loss function of the mine pressure simulation real-time calibration model obtained by training is: In the above formula, Loss represents the loss function of the mine pressure simulation real-time calibration model.

Images