Coal mine hole fracture seepage analysis and prediction system based on deep learning model

Abstract

The invention discloses a coal mine hole fracture seepage analysis and prediction system based on a deep learning model, and relates to the technical field of coal mine safety. Comprising a geological data acquisition and three-dimensional model construction module, a data preprocessing and cleaning module, a deep learning feature extraction and training module, a graph neural network optimization and topology modeling module and a seepage prediction and risk assessment module, geological exploration data and hydrogeological information of a coal mine area are obtained, and a three-dimensional geological model containing pores and fractures is constructed. Through the deep learning method combining the convolutional neural network and the graph neural network, the accuracy of coal mine fracture seepage analysis is remarkably improved. The CNN extracts local features, the GNN optimizes a global topological structure, and the seepage prediction precision is improved. The model not only can accurately predict a seepage path, but also can evaluate safety risks, help mine managers to identify hidden dangers in advance, reduce water damage and gas accumulation accidents, and optimize mine safety management.

Description

Technical Field

[0001] The present invention relates to the field of coal mine safety technology, and in particular to a coal mine pore fissure seepage analysis and prediction system based on a deep learning model. Background Art

[0002] Coal mine pore and fracture seepage analysis and prediction based on deep learning models is the process of modeling and predicting the water or air flow seepage behavior of pore and fracture systems in coal mines using deep learning technology. During coal mining, the pore and fracture structure of coal seams directly affect the seepage characteristics of groundwater or gas, which is crucial for safe production, mine ventilation, and water hazard prevention. The deep learning model automatically extracts potential nonlinear laws and predicts seepage behavior under different conditions by learning historical data (such as pore distribution, fracture connectivity, permeability, etc.). This method can provide more accurate predictions and analysis in complex coal mine environments, especially when the fracture network is complex or difficult to measure directly, to help decision makers formulate reasonable production strategies or emergency plans.

[0003] The prior art has the following deficiencies:

[0004] In the process of coal mine pore and fissure seepage analysis and prediction, deep learning models may face the problem of misjudging the topological structure of pore and fracture networks. The pore and fracture systems in coal mines usually present highly nonlinear and irregular three-dimensional distributions. The connectivity between fractures and the heterogeneity of pores have an important influence on the seepage behavior. However, during the training process, deep learning models may over-rely on local laws in historical data and ignore the global topological characteristics of the fracture system. If the model fails to accurately capture the true distribution or connectivity of fractures, it may lead to incorrect predictions of key parameters such as seepage paths and seepage velocities, which in turn may cause serious safety problems. For example, errors in seepage prediction may lead to early prediction errors of mine water hazards, or the danger of gas accumulation may be ignored, thereby threatening the safety of mine workers and causing huge economic losses. Therefore, ensuring that deep learning models can comprehensively and accurately identify and model the topological structure of fracture networks is crucial to the accuracy of seepage prediction and mine safety.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a coal mine pore fissure seepage analysis and prediction system based on a deep learning model. By combining the deep learning method of convolutional neural network (CNN) and graph neural network (GNN), the accuracy of coal mine fissure seepage analysis is significantly improved. CNN extracts local features, and GNN optimizes the global topological structure of the fracture network to comprehensively improve the accuracy of seepage prediction. This model not only provides a more accurate prediction of the seepage path, but also can evaluate safety risks, help mine managers identify potential hidden dangers in real time, take measures in advance, and effectively reduce the occurrence of safety accidents such as water damage and gas accumulation, thereby optimizing the ventilation, drainage and emergency management of the mine, and improving the reliability and stability of coal mine safety production, so as to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: A coal mine pore fissure seepage analysis and prediction system based on a deep learning model, including a geological data acquisition and three-dimensional model construction module, a data preprocessing and cleaning module, a deep learning feature extraction and training module, a graph neural network optimization and topological modeling module, and a seepage prediction and risk assessment module:

[0008] The geological data acquisition and 3D model construction module acquires geological exploration data and hydrogeological information of the coal mining area and constructs a 3D geological model including pores and fractures;

[0009] Data preprocessing and cleaning module, which preprocesses the acquired data, removes noise and incomplete information in the data, and ensures the quality and consistency of the training data;

[0010] The deep learning feature extraction and training module uses a deep learning network model to train the preprocessed data, including feature extraction and learning of the topological structure of the fracture network based on a convolutional neural network, ensuring that the model accurately identifies and accurately models the global topological characteristics of the fracture;

[0011] The graph neural network optimization and topology modeling module uses the graph neural network algorithm to optimize the connectivity modeling between fracture networks during model training. Through the relationship information between network nodes, it improves the prediction ability of the global structure of the fracture system and avoids the misjudgment of the model's excessive reliance on local data.

[0012] The seepage prediction and risk assessment module uses a regression algorithm to accurately predict the seepage path and seepage velocity based on the seepage prediction results output by the deep learning model and the actual production situation of the mine. It also outputs safety risk assessment reports under different prediction scenarios to provide decision-making support for mine safety monitoring and risk prevention and control.

[0013] Preferably, the specific steps of obtaining geological exploration data and hydrogeological information of the coal mine area and constructing a three-dimensional geological model including pores and fractures are as follows:

[0014] Firstly, relevant data are collected from geological exploration reports, drilling data and field investigation materials in the area where the coal mine is located;

[0015] Based on the collected geological exploration data, select modeling software and technical tools to construct a three-dimensional geological model;

[0016] Using the acquired geological exploration data and hydrogeological information, the data is processed through spatial interpolation technology to generate a continuous three-dimensional model;

[0017] After completing interpolation and spatial modeling, the next step is to integrate pore, fracture and hydrogeological information to form a complete three-dimensional geological model.

[0018] Preferably, the acquired data is preprocessed to remove noise and incomplete information in the data to ensure the quality and consistency of the training data. The specific steps are as follows:

[0019] After completing the collection of geological exploration data and hydrogeological information of the coal mining area, as well as the construction of a three-dimensional geological model, the acquired data is cleaned and denoised;

[0020] Process and interpolate missing data in the acquired data;

[0021] In order to ensure that different features receive balanced attention in model training, the data is standardized and normalized;

[0022] After completing data cleaning and standardization, the next step is to integrate and extract features from all the data.

[0023] Preferably, the preprocessed data is trained using a deep learning network model, which includes feature extraction and learning of the topological structure of the fracture network based on a convolutional neural network. The specific steps to ensure that the model accurately identifies and accurately models the global topological characteristics of the fracture are as follows:

[0024] After completing data preprocessing, a convolutional neural network model is constructed based on the cleaned and standardized data;

[0025] After the convolutional neural network is constructed, the preprocessed coal mine pore and fracture data are formatted to serve as the input of the convolutional neural network;

[0026] Use the preprocessed dataset to train the network;

[0027] After training, the performance of the model was evaluated to ensure that it could accurately identify the global topological characteristics of the coal mine fracture network.

[0028] Preferably, the specific steps of using the graph neural network algorithm to optimize the connectivity modeling between fracture networks and avoid the misjudgment of the model's excessive reliance on local data are as follows:

[0029] First, the coal mine fracture network is mapped into a graph structure, where the nodes of the graph represent each fracture in the coal mine, and the edges represent the connection relationship between fractures. To calculate the node embedding of the graph, the adjacency matrix of the fracture network is first constructed. The connection matrix calculation formula is:

[0030]

[0031] In the formula, A ij is the connection relationship between node i and node j. If there is connectivity between crack i and crack j, then A ij =1, otherwise 0;

[0032] Next, the features of the nodes are embedded through graph convolution operations to obtain the embedding vector of the nodes. The calculation expression is as follows:

[0033]

[0034] In the formula, is the feature vector of node i in layer l, N(i) is the set of neighbor nodes of node i, N(j) is the set of neighbor nodes of node j, W (l) is the weight matrix of the lth layer, σ(·) is the activation function, is the feature vector of node j at layer l-1, i.e., the feature representation of the previous layer;

[0035] Based on the obtained node feature vectors, the global topological structure of the fracture network is optimized to further improve the prediction ability of the fracture system. The global structure optimization formula is as follows:

[0036]

[0037] In the formula, is the feature vector of node i at layer l+1, is the feature vector of node j at layer l, α ij is the attention coefficient calculated by the graph attention mechanism, which indicates the influence of node j on node i. The calculation formula is as follows:

[0038]

[0039] In the formula, h i and h j are the feature vectors of node i and node j respectively, a is the learning parameter in the attention mechanism, T is the transpose operation of the matrix, LeakyReLU(·) is the activation function of the linear rectifier unit with leakage, and h kis the feature vector of node k in the network, and || is the concatenation operation of the feature vectors.

[0040] Preferably, based on the seepage prediction results output by the deep learning model, combined with the actual production situation of the mine, the regression algorithm is used to accurately predict the seepage path and seepage velocity, and the safety risk assessment report under different prediction scenarios is output to provide decision support for the safety monitoring and risk prevention and control of the mine. The specific steps are as follows:

[0041] First, based on the output results of the deep learning model, a seepage path prediction model was constructed, and the seepage path was described by the regression method. The formula is as follows:

[0042]

[0043] In the formula, F path (x,y,z) is the percolation path at position (x,y,z), X p is the pth input feature that affects the percolation path, β 0 is the intercept term of the regression model, β p is the pth input feature X p The relevant regression coefficient, n is the total number of input features, ∈ is the error term of the model;

[0044] Based on the prediction of the seepage path, the next step is to predict the seepage velocity. The formula is as follows:

[0045]

[0046] Where V(x,y,z) is the seepage velocity, α p is the weight coefficient in the regression model, k(x,y,z) is the permeability coefficient at the position (x,y,z), and L path (x,y,z) is the seepage path length at position (x,y,z), γ is the weight coefficient of the permeability coefficient k(x,y,z), and δ is the seepage path length L path The weight coefficient of (x, y, z), η is the percolation path F path (x, y, z) is the weight coefficient;

[0047] Finally, based on the prediction results of seepage path and seepage velocity, a safety risk assessment model for the mine was established. path (x, y, z) and seepage velocity V(x, y, z) are used to make risk prediction. The safety risk assessment calculation formula is as follows:

[0048] R(x,y,z)

[0049] =γ 0 +γ 1V(x,y,z)+γ 2 ρ crack (x,y,z)+γ 3 P(x,y,z)+γ 4 L rock (x,y,z)+γ 5 F path (x,y,z)

[0050] Where R(x,y,z) is the safety risk index at position (x,y,z), V(x,y,z) is the seepage velocity at position (x,y,z), and ρ crack (x,y,z) is the crack density at position (x,y,z), P(x,y,z) is the water level pressure at position (x,y,z), γ 0 is the constant term in the regression model, γ 1 , γ 2 , γ 3 and γ 4 are regression coefficients, γ 1 Used to quantify the impact of seepage velocity on mine safety risks, γ 2 Used to quantify the impact of crack density on safety risk, γ 3 Used to express the contribution of water level pressure to safety risk, γ 4 It is used to indicate the contribution of stratum lithology to mine safety risk, γ 5 is the influence coefficient of seepage path on safety risk assessment.

[0051] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0052] The present invention significantly improves the accuracy of coal mine fissure seepage analysis by combining the deep learning method of convolutional neural network (CNN) and graph neural network (GNN). CNN performs well in extracting local spatial characteristics of coal mine geological data, and can accurately capture the geometric shape, distribution density and relationship of the fissures with the surrounding environment, while GNN further optimizes the global topological structure modeling of the fissure network, accurately reflecting the connectivity between the fissures. Through this combination, the model can comprehensively analyze the local and global characteristics of the coal mine fissure system, eliminate the lack of adaptability of traditional methods to complex mining environments, and provide more accurate seepage predictions. The application of this model greatly improves the accuracy of coal mine seepage analysis, makes the prediction of mine seepage behavior more reliable, and provides strong technical support for mine safety.

[0053] Through the seepage prediction and safety risk assessment of the deep learning model, the present invention enables coal mine managers to identify potential safety hazards in real time and take countermeasures in advance. The model can not only accurately predict the seepage path and speed, but also evaluate the safety risks under different seepage scenarios, helping mine managers to identify high-risk areas that may cause accidents such as water damage and gas accumulation. By accurately predicting the seepage behavior of the mine, managers can optimize the ventilation and drainage systems, adjust production strategies or strengthen disaster emergency management, thereby effectively reducing the occurrence of sudden accidents and ensuring the safe operation of coal mines. This intelligent risk management model provides more scientific and accurate decision-making support for the safe production of coal mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0055] Figure 1 This is a module schematic diagram of the coal mine pore and fissure seepage analysis and prediction system based on the deep learning model of the present invention. DETAILED DESCRIPTION

[0056] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0057] The present invention provides Figure 1 The coal mine pore fissure seepage analysis and prediction system based on deep learning model shown in the figure includes geological data acquisition and three-dimensional model construction module, data preprocessing and cleaning module, deep learning feature extraction and training module, graph neural network optimization and topology modeling module and seepage prediction and risk assessment module:

[0058] The geological data acquisition and 3D model construction module acquires geological exploration data and hydrogeological information of the coal mining area and constructs a 3D geological model including pores and fractures;

[0059] The specific steps to obtain geological exploration data and hydrogeological information of the coal mining area and construct a three-dimensional geological model including pores and fractures are as follows:

[0060] Firstly, relevant data are collected from geological exploration reports, drilling data and field investigation materials in the area where the coal mine is located;

[0061] Geological exploration data usually include the thickness, lithology, porosity, distribution of fractures, burial depth of coal seams, etc. These data are obtained through geological exploration drilling and physical survey technology, which can provide detailed information about the underground structure of the mining area. At the same time, hydrogeological data are also crucial, including groundwater level, flow direction, permeability and water quality. These data help to understand the behavior of groundwater in the coal mining area and its interaction with pore and fracture systems. In the process of data collection, the representativeness and accuracy of the data must be ensured in order to provide reliable basic information for the three-dimensional geological model.

[0062] Based on the collected geological exploration data, select modeling software and technical tools to construct a three-dimensional geological model;

[0063] Common 3D modeling tools include GOCAD, Petrel, GeoStudio, etc. These tools can convert 2D drilling and exploration data into 3D geological models. Through interpolation algorithms, the data of drilling points are used to perform 3D modeling of the geological space of the mining area, accurately reflecting the spatial distribution and geometric morphology of pores and fractures in the mining area. The modeling of fracture networks requires special attention because fracture systems are usually irregular and their connectivity directly affects the seepage characteristics. In the modeling process, the integration of hydrogeological information also needs to be considered to ensure the consistency of groundwater flow characteristics and rock permeability.

[0064] Using the acquired geological exploration data and hydrogeological information, the data is processed through spatial interpolation technology to generate a continuous three-dimensional model;

[0065] Interpolation methods such as Kriging interpolation and inverse distance weighted (IDW) can extend limited borehole data to other areas of the mining area. For the spatial distribution of fractures and pores, interpolation can help construct the fracture network and porosity changes in the coal mining area. These models not only reflect the spatial distribution of geological and hydrological conditions, but also reflect the heterogeneity of fractures and pores (such as the density of fractures and the heterogeneity of pores). This three-dimensional spatial distribution provides very important input information for subsequent seepage analysis, especially for complex underground fracture networks. Accurate spatial modeling can effectively help understand the flow path of groundwater or airflow.

[0066] After completing interpolation and spatial modeling, the next step is to integrate pore, fracture, and hydrogeological information to form a complete three-dimensional geological model;

[0067] This process requires the fusion of lithology, porosity, fracture density, hydrogeological data and rock formation information to ensure the accuracy and completeness of the model. For fracture systems, special attention should be paid to factors such as fracture morphology, strike, inclination, and distribution density when modeling, which all affect seepage behavior. On this basis, the flow path and groundwater permeability area in the mining area can be further constructed to provide an accurate geological model for seepage analysis. The final three-dimensional model can be used for subsequent deep learning training and seepage prediction, and provide strong support for safe production and water hazard prevention in coal mines.

[0068] Data preprocessing and cleaning module, which preprocesses the acquired data, removes noise and incomplete information in the data, and ensures the quality and consistency of the training data;

[0069] The specific steps for preprocessing the acquired data, removing noise and incomplete information in the data, and ensuring the quality and consistency of the training data are as follows:

[0070] After completing the collection of geological exploration data and hydrogeological information of the coal mining area, as well as the construction of a three-dimensional geological model, the acquired data is cleaned and denoised;

[0071] First, remove the noise data that does not conform to reality by checking the outliers, duplicate data and erroneous records in the original data. For example, erroneous readings in borehole data, inconsistencies in lithology records or measurement errors in geological exploration instruments will affect the accuracy of subsequent models. In this process, common data cleaning methods include filtering of numerical ranges, correction of extreme values ​​and smoothing. For hydrogeological data, if there are abnormal groundwater level records and flow data that do not conform to the laws of the geological environment, calibration and correction are also required. The processing at this stage is to ensure the accuracy of the data and avoid noise interference with subsequent model training.

[0072] Process and interpolate missing data in the acquired data;

[0073] In coal mine exploration data, especially in the spatial distribution of underground pores and fractures, some data may be missing or incomplete. Due to the complex geological structure of coal mines, the distribution of drilling points during the exploration process may be uneven, resulting in missing data in some areas. At this time, these missing data must be processed. Commonly used methods include interpolation (such as linear interpolation, Kriging interpolation, or inverse distance weighted interpolation, etc.), which predict missing values ​​through existing data points. For example, if some data such as fracture depth and porosity are missing in some boreholes, the data of adjacent boreholes can be used for interpolation and supplementation, thereby maintaining the consistency and integrity of the data. The key to interpolation is to select a suitable interpolation method, judge the data interpolation effect at different locations according to geological characteristics, and avoid errors introduced by improper interpolation methods.

[0074] In order to ensure that different features receive balanced attention in model training, the data is standardized and normalized;

[0075] When conducting three-dimensional geological modeling, the dimensions of pores, fractures, and hydrogeological data in coal mining areas are often different. For example, the porosity may range from 0 to 1, while the fracture depth may range from a few meters to tens of meters. In order to ensure that different features receive balanced attention in model training, the data needs to be standardized and normalized. Standardization usually converts the data into a distribution with a mean of 0 and a variance of 1 by subtracting the mean and dividing by the standard deviation; normalization scales the data to a fixed range (such as 0 to 1). These two methods can eliminate the unbalanced effects of data with different dimensions on the model during training, ensure that different data features have the same weight during model training, and thus improve the stability and accuracy of the training results.

[0076] After completing data cleaning and standardization, the next step is to integrate and extract features from all the data;

[0077] For the three-dimensional geological model of coal mine pores and fissures, the data needs to be processed in spatial coordinates, and the data of each borehole needs to be associated with its corresponding geographic coordinate position to form a complete spatial data set. At the same time, feature extraction is crucial for deep learning models, and it is necessary to calculate the input features in the model, such as the density of pores, the connectivity of fractures, the lithology of each stratum, and the permeability of groundwater flow. Through these feature extractions, the deep learning model can understand the inherent relationship of the data and learn effectively. At this stage, it is also necessary to consider the fusion of multiple sources of data, such as fusing exploration data with other related data such as hydrogeological information and seismic wave data, so that the final data set is more comprehensive and can better support the deep learning model to learn and predict the seepage characteristics of coal mines.

[0078] The deep learning feature extraction and training module uses a deep learning network model to train the preprocessed data, including feature extraction and learning of the topological structure of the fracture network based on a convolutional neural network, ensuring that the model accurately identifies and accurately models the global topological characteristics of the fracture;

[0079] The preprocessed data is trained using a deep learning network model, which includes feature extraction and learning of the topological structure of the fracture network based on a convolutional neural network. The specific steps to ensure that the model accurately identifies and accurately models the global topological characteristics of the fracture are as follows:

[0080] After completing data preprocessing, a convolutional neural network model is constructed based on the cleaned and standardized data;

[0081] Convolutional neural networks are very suitable for processing data with spatial characteristics, such as the three-dimensional spatial distribution data of coal mine pores and cracks. First, design an appropriate network structure, including multiple convolutional layers, pooling layers, and fully connected layers. The role of the convolutional layer is to extract local features in the data through filters, and the pooling layer is used to reduce the dimension of the data, reduce the computational complexity, avoid overfitting, and retain the main information of the data. When designing the model, considering the complexity of the coal mine crack network, the size and depth of the convolution kernel need to be optimized so that the model can capture the geometry, direction, density and other characteristics of the cracks. In addition, the model also needs to be able to process three-dimensional data, so the design of the convolutional layer and the pooling layer must be able to process three-dimensional spatial information.

[0082] After the convolutional neural network is constructed, the preprocessed coal mine pore and fracture data are formatted to serve as the input of the convolutional neural network;

[0083] The three-dimensional geological data of coal mines are usually represented in the form of grids or voxels, representing the three-dimensional space of the coal mine area. Before the data is input into the CNN model, it is necessary to convert the data into an appropriate input format, such as encoding each drilling data, the depth and position of the crack into a vector or matrix form suitable for model processing. The convolutional neural network automatically extracts local features in the data (such as the size, shape, density, etc. of the cracks) through the convolution layer. Due to the highly nonlinear characteristics of the distribution of coal mine cracks, the convolutional neural network can gradually extract from low-level features to more complex global features through multi-layer convolution operations, thereby accurately capturing the topological structure of the cracks.

[0084] Use the preprocessed dataset to train the network;

[0085] The weights of the network are updated through backpropagation and optimization algorithms (such as Adam or SGD) so that the model can accurately predict the topological structure of the coal mine fracture network. During training, the data set is first divided into a training set and a validation set to monitor the generalization ability of the model during the training process and prevent overfitting. During the training process, the model continuously adjusts the weights so that the network can maximize the spatial characteristics of the fractures and learn the topological relationship of the fracture network. For example, the network may identify which areas have denser fractures and which areas have higher connectivity between fractures, which is very critical for subsequent seepage analysis. In addition, a suitable loss function needs to be set during the training process to measure the prediction error of the model. Commonly used loss functions include mean square error (MSE) or cross-entropy.

[0086] After training, the performance of the model was evaluated to ensure that it could accurately identify the global topological characteristics of the coal mine fracture network;

[0087] In the evaluation phase, the prediction accuracy of the model is tested by comparing the prediction results of the test set with the actual observed data. Various performance indicators can be used for evaluation, such as accuracy, recall, F1 score, etc. If the performance of the model does not meet expectations, the model can be further optimized by adjusting the network structure, increasing the amount of data, or changing the training strategy (for example, adjusting the learning rate or using a stronger regularization method). Once the evaluation is passed, the model can be applied to actual coal mine seepage analysis to help analyze the connectivity of the fracture network, seepage paths, seepage velocity, etc., and provide accurate data support for coal mine safety monitoring and risk assessment.

[0088] The graph neural network optimization and topology modeling module uses the graph neural network algorithm to optimize the connectivity modeling between fracture networks during model training. Through the relationship information between network nodes, it improves the prediction ability of the global structure of the fracture system and avoids the misjudgment of the model's excessive reliance on local data.

[0089] The specific steps to use the graph neural network algorithm to optimize the connectivity modeling between fracture networks and avoid the misjudgment of the model's over-reliance on local data are as follows:

[0090] First, the coal mine fracture network is mapped into a graph structure, where the nodes of the graph represent each fracture in the coal mine, and the edges represent the connection relationship between fractures. To calculate the node embedding of the graph, the adjacency matrix of the fracture network is first constructed. This matrix represents the connectivity between fractures. The connection matrix calculation formula is:

[0091]

[0092] In the formula, A ij is the connection relationship between node i and node j. If there is connectivity between fracture i and fracture j (for example, the fractures are adjacent or there is a water flow channel), then A ij =1, otherwise 0;

[0093] Next, the features of the nodes are embedded through graph convolution operations to obtain the embedding vector of the nodes. The calculation expression is as follows:

[0094]

[0095] In the formula, is the feature vector of node i in layer l, N(i) is the set of neighbor nodes of node i, N(j) is the set of neighbor nodes of node j, W (l) is the weight matrix of the lth layer, which is used to linearly transform the features of neighbor nodes. σ(·) is the activation function (such as ReLU) used to introduce nonlinearity. is the feature vector of node j at layer l-1, i.e., the feature representation of the previous layer;

[0096] As the number of graph convolution layers increases, the features of the nodes will gradually combine the information of more neighboring nodes, thereby obtaining richer global topological information. These embedded vectors will serve as important inputs for subsequent model training, conveying global structural information between cracks.

[0097] Based on the obtained node feature vectors, the global topological structure of the fracture network is optimized to further improve the prediction ability of the fracture system. In order to ensure that the model does not rely too much on local data, we need to aggregate and optimize information at the global level. Specifically, the graph attention mechanism (GAT) is used to dynamically adjust the influence of each neighbor node, so as to more effectively aggregate the information of neighbor nodes and improve the learning ability of the global structure. The global structure optimization formula is as follows:

[0098]

[0099] In the formula, is the feature vector of node i at layer l+1, is the feature vector of node j at layer l, α ij is the attention coefficient calculated by the graph attention mechanism, which indicates the influence of node j on node i. The calculation formula is as follows:

[0100]

[0101] In the formula, h i and h j are the feature vectors of node i and node j respectively, a is the learning parameter in the attention mechanism, indicating the importance of the relationship between node features, T is the transpose operation of the matrix, LeakyReLU(·) is the linear rectifier unit activation function with leakage, and h k is the feature vector of node k in the network, ∥ is the concatenation operation of feature vectors, which is used to connect the feature vectors of node i and node j to generate a new feature pair.

[0102] In the above steps, by calculating the attention coefficient α of each neighbor node ij , we can dynamically adjust the contribution of neighbor nodes to the embedding of the target node, further improving the learning effect of the global topology. This process helps to optimize the connectivity modeling of the fracture network, so that the deep learning model can better capture the global relationship between fractures when predicting the percolation behavior of the fracture system, rather than relying solely on local information.

[0103] The seepage prediction and risk assessment module uses a regression algorithm to accurately predict the seepage path and seepage velocity based on the seepage prediction results output by the deep learning model and the actual production situation of the mine. It also outputs safety risk assessment reports under different prediction scenarios to provide decision-making support for safety monitoring and risk prevention and control of mines.

[0104] Based on the seepage prediction results output by the deep learning model and combined with the actual production situation of the mine, the regression algorithm is used to accurately predict the seepage path and seepage velocity, and the safety risk assessment report under different prediction scenarios is output to provide decision support for the safety monitoring and risk prevention and control of the mine. The specific steps are as follows:

[0105] First, based on the output results of the deep learning model, a seepage path prediction model was constructed, and the seepage path was described by the regression method. The formula is as follows:

[0106]

[0107] In the formula, F path (x,y,z) is the percolation path at position (x,y,z), X p is the pth input feature that affects the seepage path (such as porosity, fracture distribution, lithology, etc.), β 0 is the intercept term of the regression model, which means that among all the independent variables {X p , X 1 , X 2 ,……,X n} are all 0, the basic value of the seepage path, β p is the pth input feature X p The relevant regression coefficient, n is the total number of input features, ∈ is the error term of the model;

[0108] The goal of this step is to obtain accurate predictions of the seepage path and provide input data for subsequent calculations of the seepage velocity.

[0109] Based on the prediction of the seepage path, the next step is to predict the seepage velocity. The prediction of the seepage velocity depends not only on the permeability coefficient and geometric length of the fracture, but also on the geometric characteristics of the seepage path. The formula is as follows:

[0110]

[0111] Where V(x,y,z) is the seepage velocity, α p It is the weight coefficient in the regression model, which is used to represent each input feature X p The contribution to the seepage velocity V(x,y,z), k(x,y,z) is the permeability coefficient at the position (x,y,z), L path (x,y,z) is the seepage path length at position (x,y,z), γ is the weight coefficient of the permeability coefficient k(x,y,z), and δ is the seepage path length L path The weight coefficient of (x, y, z), η is the percolation path F path (x, y, z) is the weight coefficient;

[0112] Finally, based on the prediction results of seepage path and seepage velocity, a safety risk assessment model for the mine was established. path (x, y, z) and seepage velocity V(x, y, z) are used to make risk prediction. The safety risk assessment calculation formula is as follows:

[0113] R(x,y,z)

[0114] =γ 0 +γ 1 V(x,y,z)+γ 2 ρ crack (x,y,z)+γ 3 P(x,y,z)+γ 4 L rock (x,y,z)+γ 5 F path (x,y,z)

[0115] , where R(x,y,z) is the safety risk index at position (x,y,z), V(x,y,z) is the seepage velocity at position (x,y,z), and ρ crack (x,y,z) is the crack density at position (x,y,z), P(x,y,z) is the water level pressure at position (x,y,z), γ 0 is the constant term (intercept) in the regression model, representing the baseline effect of all input variables (i.e., seepage velocity, fracture density, etc.), and is usually used to adjust the initial prediction value of the model. 1 , γ 2 , γ 3 and γ 4 are regression coefficients, γ 1 Used to quantify the impact of seepage velocity on mine safety risks, γ 2 Used to quantify the impact of crack density on safety risk, γ 3 Used to express the contribution of water level pressure to safety risk, γ 4 It is used to indicate the contribution of stratum lithology to mine safety risk, γ 5 is the influence coefficient of seepage path on safety risk assessment.

[0116] Implementation method 1: This implementation method solves the complex problems in modeling the topological structure of coal mine fracture networks by combining two deep learning models, namely convolutional neural network (CNN) and graph neural network (GNN). Coal mine fracture networks usually have high nonlinearity, heterogeneity and complex spatial structure. Traditional modeling methods cannot fully capture these characteristics, while convolutional neural network (CNN) and graph neural network (GNN) can effectively handle these problems, thereby providing more accurate analysis for coal mine seepage prediction.

[0117] First, convolutional neural networks are used to extract features from the three-dimensional geological data of coal mines. In coal mine fracture modeling, the geometry, size, distribution density and spatial relationship between fractures all have an important influence on the seepage behavior. Convolutional neural networks (CNNs) are very suitable for processing such spatial data due to their powerful spatial feature learning capabilities. Through the convolution layer, CNN can extract local fracture features from the input three-dimensional data, such as fracture depth, length, distribution pattern, etc.

[0118] At this stage, the original coal mine data, such as drilling data, fracture distribution, porosity, etc., are abstracted layer by layer through the multi-layer convolution operation of the convolutional neural network to extract spatial features of different scales. These convolution operations can not only extract low-level features (such as the shape and distribution of fractures), but also gradually improve the network's understanding of the coal mine fracture system through multi-level feature learning. In this way, CNN can capture subtle changes in the fracture system, which is crucial for the accurate prediction of subsequent seepage behavior.

[0119] Due to the complexity of coal mine fissures, local feature extraction relying solely on convolutional neural networks cannot fully describe the global connectivity between fissures. In this case, graph neural networks (GNNs) can help model the complex connectivity between fissures. Graph neural networks can treat coal mine fissures as nodes in a graph, and the connectivity between fissures as edges connecting these nodes. Each node represents a feature of a fissure (such as fissure depth, porosity, etc.), while the edges of the graph represent the physical connectivity between the fissures.

[0120] During the model training process, GNN transmits and updates information between nodes through graph convolution operations. Each node receives information about its neighboring nodes through the graph convolution layer, thereby capturing the global topological structure in the fracture network. In this way, the graph neural network can learn the connectivity between fractures on a global scale, improve the understanding of the coal mine fracture network, thereby reducing the model's dependence on local data and avoiding local misjudgments that may occur in traditional convolutional neural networks in fracture network modeling.

[0121] The combination of convolutional neural networks and graph neural networks is the core advantage of this implementation. By combining CNN and GNN, the local features and global topological characteristics of the cracks can be integrated, thereby improving the modeling accuracy of the coal mine crack network. In the actual training process, CNN is first used to extract features from coal mine data to obtain local information about the cracks, and then this information is used as input and further passed to the graph neural network for global connectivity modeling. Through this joint training method, CNN is responsible for learning the geometric and spatial characteristics of the cracks, while GNN helps capture the global structure of the crack system. The two complement each other and jointly optimize the performance of the model.

[0122] In addition, joint training can also improve the generalization ability of the model and reduce the risk of overfitting. Since the convolutional neural network can extract effective local features from a large amount of data, and the graph neural network can strengthen the model's learning of the global structure of the fracture network, the overall performance of the model has been effectively improved. This deep fusion method has shown strong predictive capabilities for complex coal mine fracture seepage analysis, especially when faced with unconventional fracture distribution or abnormal seepage paths.

[0123] Finally, the jointly trained deep learning model can be used for the prediction and safety assessment of coal mine seepage. At this stage, the model processed by CNN and GNN outputs information such as the seepage path, seepage velocity, and potential water hazard risk of the fracture. These prediction results can provide important references for the safe production of coal mines. For example, when it is found that the connectivity of the fracture system is poor or the seepage path is abnormal, preventive measures can be taken in advance to avoid safety accidents such as water damage or gas accumulation.

[0124] In addition, the model can also evaluate the seepage risk in different scenarios, help mine managers formulate scientific safety strategies, and optimize the ventilation and water hazard prevention measures of the mine. Through real-time monitoring and prediction, the safety of coal mines has been greatly improved, thereby effectively reducing the risk of accidents during mine operations.

[0125] Implementation method 2: This implementation method combines different types of coal mine data (such as geological exploration data, hydrogeological information, and actual measurement data) through multi-source data fusion technology to provide a more comprehensive input for the deep learning model. This fusion of multi-source data can provide more accurate predictions for fracture seepage analysis, thereby improving the safety and production efficiency of coal mines.

[0126] In the process of multi-source data fusion in coal mines, it is first necessary to collect diversified data from different fields, including geological exploration data (such as drilling data, fracture depth, porosity, etc.), hydrogeological information (such as groundwater level, flow rate, permeability, etc.), and actual measurement data (such as real-time mine water level, gas concentration, etc.). These data can fully reflect the underground hydrological environment and fracture structure of coal mines, and provide sufficient information for deep learning model training.

[0127] The collected data usually have different time scales, spatial scales and data formats, so these data must be standardized and processed uniformly. For example, data from different sources are preprocessed through interpolation, data cleaning, denoising and missing value filling to ensure data accuracy and consistency. At the same time, through data standardization and normalization, all types of data have the same dimension, thus avoiding some data from occupying too much weight in model training.

[0128] After data preprocessing, the next step is joint training of deep learning models. By combining convolutional neural networks (CNN) with recurrent neural networks (RNN), the model can simultaneously process spatial features and temporal dynamic information in coal mine data. CNN is used to extract spatial features in coal mine geological data, especially information such as the geometry, distribution, and density of fractures. Through multi-layer convolution operations, CNN can extract local spatial features of the fracture network and provide necessary input for subsequent analysis.

[0129] Recurrent neural networks (RNNs) are used to capture the temporal characteristics of hydrogeological information. For example, data such as groundwater levels and flow rates vary over time, and RNNs can process these time series data to help the model capture the flow characteristics of groundwater at different time points. In addition, combined with long short-term memory (LSTM) networks, the model can effectively handle data changes over long time spans, making the model's prediction of the hydrological environment more accurate.

[0130] During the joint training process, the model will learn how to process spatial and temporal data simultaneously to extract comprehensive characteristics of fracture seepage. Through the input data in the training dataset, the model will gradually optimize its weights, thereby improving the prediction ability of coal mine fracture networks and seepage behavior. During the training process, the model is evaluated by cross-validation method, and the prediction error of the model is measured by loss function (such as mean square error).

[0131] After training is completed, the accuracy of the model is verified through the test set. If the prediction results of the model meet expectations, it can be applied to actual coal mine seepage analysis to provide support for safe production in coal mines. If there are deviations in the prediction results, the model performance can be further optimized by adjusting the network structure or increasing the amount of data. In addition, through ensemble learning technology, the prediction results of multiple models can be combined to further improve the stability and accuracy of the prediction.

[0132] Ultimately, the jointly trained deep learning model will be used for real-time prediction of coal mine fissure seepage. The model can provide predictions of key parameters such as seepage path and seepage velocity, and

[0133] Based on these prediction results, a safety risk assessment report for the mine is generated. By predicting seepage under different scenarios, the model can effectively identify potential safety hazards and provide decision support for mine management.

[0134] Through this method of multi-source data fusion and deep learning joint training, the seepage prediction of coal mines can not only take into account spatial characteristics, but also comprehensively consider temporal change factors, thereby providing more accurate analysis and judgment for mine safety management and reducing the risk of disasters such as water damage and gas accumulation.

[0135] Implementation method 3: This implementation method solves the spatial resolution problem in coal mine fissure seepage analysis by introducing multi-scale analysis technology. Due to the large differences in the spatial scales of coal mine fissures, the traditional unified scale modeling method may ignore some details, resulting in inaccurate prediction results. Through multi-scale analysis, different spatial resolution modeling can be adopted according to the characteristics of different regions, thereby improving the accuracy and reliability of coal mine seepage prediction.

[0136] Before starting multi-scale modeling, the coal mine area is first spatially divided. According to the geological characteristics and the distribution of fractures, the coal mine area is divided into different scale levels. For example, high-resolution grids are used for modeling in areas with dense fractures, while lower-resolution grids are used for modeling in areas with sparse fractures. This spatial division can dynamically adjust the modeling accuracy according to the complexity of different areas, so that the model can fully reflect the seepage characteristics of the coal mine at different scales.

[0137] In the data preprocessing stage, the data of different scale areas will be normalized to ensure that the data at different scales have the same dimension and are compatible. The data details of the high-resolution area will be retained to accurately reflect the fine structure of the crack, while the data of the low-resolution area will be interpolated and smoothed to ensure global consistency.

[0138] After completing data preprocessing and spatial scale division, a multi-scale convolutional neural network (MS-CNN) is used for model training. MS-CNN can process data at different scales simultaneously to extract global and local features. Through multi-layer convolution operations, the model can extract global features in low-resolution areas and more detailed crack structure features in high-resolution areas.

[0139] The MS-CNN model can fully learn the different scale characteristics of coal mine fissures by performing parallel processing at multiple scales. During the training process, CNN can automatically adapt to the resolution requirements of different regions, effectively reducing the computational overhead in high-resolution regions, while avoiding the loss of key details in low-resolution regions, thereby ensuring the accuracy of seepage prediction.

[0140] The multi-scale trained model will be used to predict the seepage path and seepage velocity in coal mines. In this process, the MS-CNN model not only takes into account the characteristics of different scale areas, but also combines the specific geological environment of the coal mine area, which can provide more accurate seepage prediction results. Through the regression algorithm, the seepage path and velocity are optimized to improve the accuracy of the prediction results.

[0141] In addition, based on the seepage prediction results, risk assessment can be carried out to analyze the seepage risks in different areas and provide decision support for mine management. This method can effectively solve the accuracy problem of traditional models in complex mining areas and improve the reliability and operability of coal mine seepage prediction.

[0142] Through the above-mentioned multi-scale analysis method, the seepage prediction of coal mines can not only obtain high-precision results at different spatial scales, but also help mine managers take more reasonable safety management measures for different areas to ensure safe production in mines.

[0143] The present invention significantly improves the accuracy of coal mine fissure seepage analysis by combining the deep learning method of convolutional neural network (CNN) and graph neural network (GNN). CNN performs well in extracting local spatial characteristics of coal mine geological data, and can accurately capture the geometric shape, distribution density and relationship of the fissures with the surrounding environment, while GNN further optimizes the global topological structure modeling of the fissure network, accurately reflecting the connectivity between the fissures. Through this combination, the model can comprehensively analyze the local and global characteristics of the coal mine fissure system, eliminate the lack of adaptability of traditional methods to complex mining environments, and provide more accurate seepage predictions. The application of this model greatly improves the accuracy of coal mine seepage analysis, makes the prediction of mine seepage behavior more reliable, and provides strong technical support for mine safety.

[0144] Through the seepage prediction and safety risk assessment of the deep learning model, the present invention enables coal mine managers to identify potential safety hazards in real time and take countermeasures in advance. The model can not only accurately predict the seepage path and speed, but also evaluate the safety risks under different seepage scenarios, helping mine managers to identify high-risk areas that may cause accidents such as water damage and gas accumulation. By accurately predicting the seepage behavior of the mine, managers can optimize the ventilation and drainage systems, adjust production strategies or strengthen disaster emergency management, thereby effectively reducing the occurrence of sudden accidents and ensuring the safe operation of coal mines. This intelligent risk management model provides more scientific and accurate decision-making support for the safe production of coal mines.

[0145] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A coal mine pore fissure seepage analysis and prediction system based on a deep learning model, characterized in that: It includes geological data acquisition and 3D model construction module, data preprocessing and cleaning module, deep learning feature extraction and training module, graph neural network optimization and topology modeling module, and seepage prediction and risk assessment module: The geological data acquisition and 3D model construction module acquires geological exploration data and hydrogeological information of the coal mining area and constructs a 3D geological model including pores and fractures; Data preprocessing and cleaning module, which preprocesses the acquired data, removes noise and incomplete information in the data, and ensures the quality and consistency of the training data; The deep learning feature extraction and training module uses a deep learning network model to train the preprocessed data, including feature extraction and learning of the topological structure of the fracture network based on a convolutional neural network, ensuring that the model accurately identifies and accurately models the global topological characteristics of the fracture; The graph neural network optimization and topology modeling module uses the graph neural network algorithm to optimize the connectivity modeling between fracture networks during model training. Through the relationship information between network nodes, it improves the prediction ability of the global structure of the fracture system and avoids the misjudgment of the model's excessive reliance on local data. The seepage prediction and risk assessment module uses a regression algorithm to accurately predict the seepage path and seepage velocity based on the seepage prediction results output by the deep learning model and the actual production situation of the mine. It also outputs safety risk assessment reports under different prediction scenarios to provide decision-making support for mine safety monitoring and risk prevention and control.

2. The coal mine pore fissure seepage analysis and prediction system based on deep learning model according to claim 1 is characterized in that: The specific steps to obtain geological exploration data and hydrogeological information of the coal mining area and construct a three-dimensional geological model including pores and fractures are as follows: Firstly, relevant data are collected from geological exploration reports, drilling data and field investigation materials in the area where the coal mine is located; Based on the collected geological exploration data, select modeling software and technical tools to construct a three-dimensional geological model; Using the acquired geological exploration data and hydrogeological information, the data is processed through spatial interpolation technology to generate a continuous three-dimensional model; After completing interpolation and spatial modeling, the next step is to integrate pore, fracture and hydrogeological information to form a complete three-dimensional geological model.

3. The coal mine pore fissure seepage analysis and prediction system based on deep learning model according to claim 1 is characterized in that: The specific steps for preprocessing the acquired data, removing noise and incomplete information in the data, and ensuring the quality and consistency of the training data are as follows: After completing the collection of geological exploration data and hydrogeological information of the coal mining area, as well as the construction of a three-dimensional geological model, the acquired data is cleaned and denoised; Process and interpolate missing data in the acquired data; In order to ensure that different features receive balanced attention in model training, the data is standardized and normalized; After completing data cleaning and standardization, the next step is to integrate and extract features from all the data.

4. The coal mine pore fissure seepage analysis and prediction system based on deep learning model according to claim 1, characterized in that: The preprocessed data is trained using a deep learning network model, which includes feature extraction and learning of the topological structure of the fracture network based on a convolutional neural network. The specific steps to ensure that the model accurately identifies and accurately models the global topological characteristics of the fracture are as follows: After completing data preprocessing, a convolutional neural network model is constructed based on the cleaned and standardized data; After the convolutional neural network is constructed, the preprocessed coal mine pore and fracture data are formatted to serve as the input of the convolutional neural network; Use the preprocessed dataset to train the network; After training, the performance of the model was evaluated to ensure that it could accurately identify the global topological characteristics of the coal mine fracture network.

5. The coal mine pore fissure seepage analysis and prediction system based on deep learning model according to claim 1, characterized in that: The specific steps to use the graph neural network algorithm to optimize the connectivity modeling between fracture networks and avoid the misjudgment of the model's over-reliance on local data are as follows: First, the coal mine fracture network is mapped into a graph structure, where the nodes of the graph represent each fracture in the coal mine, and the edges represent the connection relationship between fractures. To calculate the node embedding of the graph, the adjacency matrix of the fracture network is first constructed. The connection matrix calculation formula is: In the formula, A ij is the connection relationship between node i and node j. If there is connectivity between crack i and crack j, then A ij =1, otherwise 0; Next, the features of the nodes are embedded through graph convolution operations to obtain the embedding vector of the nodes. The calculation expression is as follows: In the formula, is the feature vector of node i in layer l, N(i) is the set of neighbor nodes of node i, N(j) is the set of neighbor nodes of node j, W (l) is the weight matrix of the lth layer, σ(·) is the activation function, is the feature vector of node j at layer l-1, i.e., the feature representation of the previous layer; Based on the obtained node feature vectors, the global topological structure of the fracture network is optimized to further improve the prediction ability of the fracture system. The global structure optimization formula is as follows: α ij is the attention coefficient calculated by the graph attention mechanism, which indicates the influence of node j on node i. The calculation formula is as follows: In the formula, h i and h j are the feature vectors of node i and node j respectively, a is the learning parameter in the attention mechanism, T is the transpose operation of the matrix, LeakyReLU(·) is the activation function of the linear rectifier unit with leakage, and h k is the feature vector of node k in the network, and || is the concatenation operation of the feature vectors.

6. The coal mine pore fissure seepage analysis and prediction system based on deep learning model according to claim 1, characterized in that: Based on the seepage prediction results output by the deep learning model and combined with the actual production situation of the mine, the regression algorithm is used to accurately predict the seepage path and seepage velocity, and the safety risk assessment report under different prediction scenarios is output to provide decision support for the safety monitoring and risk prevention and control of the mine. The specific steps are as follows: First, based on the output results of the deep learning model, a seepage path prediction model was constructed, and the seepage path was described by the regression method. The formula is as follows: In the formula, F path (x,y,z) is the percolation path at position (x,y,z), X p is the pth input feature that affects the percolation path, β0 is the intercept term of the regression model, and β p is the pth input feature X p The relevant regression coefficient, n is the total number of input features, ∈ is the error term of the model; Based on the prediction of the seepage path, the next step is to predict the seepage velocity. The formula is as follows: Where V(x,y,z) is the seepage velocity, α p is the weight coefficient in the regression model, k(x,y,z) is the permeability coefficient at the position (x,y,z), and L path (x,y,z) is the seepage path length at position (x,y,z), γ is the weight coefficient of the permeability coefficient k(x,y,z), and δ is the seepage path length L path The weight coefficient of (x, y, z), η is the percolation path F path (x, y, z) is the weight coefficient; Finally, based on the prediction results of seepage path and seepage velocity, a safety risk assessment model for the mine was established. path (x, y, z) and seepage velocity V(x, y, z) are used to make risk prediction. The safety risk assessment calculation formula is as follows: R(x,y,z) =γ0+γ1V(x,y,z)+γ2ρ crack (x,y,z)+γ3P(x,y,z)+γ4L rock (x,y,z)+γ5F path (x,y,z) Where R(x,y,z) is the safety risk index at position (x,y,z), V(x,y,z) is the seepage velocity at position (x,y,z), and ρ crack (x,y,z) is the fracture density at position (x,y,z), P(x,y,z) is the water level pressure at position (x,y,z), γ0 is the constant term in the regression model, γ1, γ2, γ3 and γ4 are all regression coefficients, γ1 is used to quantify the impact of seepage velocity on mine safety risk, γ2 is used to quantify the impact of fracture density on safety risk, γ3 is used to indicate the contribution of water level pressure to safety risk, γ4 is used to indicate the contribution of stratum lithology to mine safety risk, and γ5 is the influence coefficient of seepage path on safety risk assessment.

Images