Geological constraint dynamic fusion-based low-permeability coal seam gas content accurate prediction method

Abstract

The invention discloses a method for accurately predicting the gas content of a low-permeability coal seam based on geological constraint dynamic fusion. The method comprises the following steps: firstly, constructing a multi-source data system comprising four categories of basic geology, geophysical logging, geologic structure and ground stress and coal body structure, and carrying out standardization processing; secondly, constructing a sample set by taking the actually measured gas content as a label, and screening out an optimal feature subset through a dynamic feature fusion mechanism of geological mechanism primary screening combined with model-driven fine screening; performing training and verification by utilizing machine learning models such as XGBoost and the like, and performing gridding interpolation and prediction on a target area after the precision is reached, so as to generate a gas content distribution diagram. According to the method, multi-source data are fused, a geological constraint mechanism is embedded, the problems that a traditional method is single in data source, ignores geological causes and is poor in adaptability are solved, accurate description of gas occurrence space heterogeneity can be achieved, prediction accuracy is improved, and a reliable basis is provided for gas extraction design and disaster early warning.

Description

Technical Field

[0001] This application belongs to the field of gas content prediction technology, specifically involving a method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints. Background Technology

[0002] In coal seam gas disaster prevention and gas extraction and utilization, gas content is a core basic parameter, and its prediction accuracy directly determines the safety production efficiency and gas resource development benefits of coal mines. Currently, the mainstream technologies for predicting gas content in low-permeability coal seams are mainly divided into two categories. One category is prediction methods based on geological exploration data. This method uses core sampling and laboratory desorption (direct method) to determine the gas content at discrete borehole points. Combined with basic geological parameters such as coal seam depth, coal seam thickness, and roof and floor lithology, it utilizes methods such as multiple regression analysis and ordinary kriging interpolation to achieve spatial extrapolation prediction of gas content across the entire mining area. The other category is prediction methods based on geophysical logging data. This method uses conventional logging curves such as sonic transit time, density, resistivity, and natural gamma rays, and establishes the correlation between logging response and measured gas content through empirical formulas or statistical models to continuously estimate gas content in sections without core sampling.

[0003] Given the inherent heterogeneity of low-permeability coal seams and the fact that gas occurrence is controlled by multiple geological factors, existing technologies suffer from the following inherent defects, leading to insufficient prediction accuracy: Single data source and lack of information dimensions: relying solely on geological exploration or well logging data without integrating key gas-controlling factors such as geological structure, geostress, and coal body structure, resulting in insufficient data support; Neglecting geological constraint mechanisms: prediction models do not incorporate the geological mechanisms of gas occurrence, relying only on statistical methods to establish correlations, failing to reflect the control of tectonic deformation and geostress fields on gas adsorption-desorption-migration. The model is static and lacks adaptability: it uses fixed feature combinations or empirical formulas, which cannot dynamically adjust the input parameters according to the geological conditions of different mining areas, resulting in insufficient adaptability to the spatial heterogeneity of low-permeability coal seams; the "point-surface" scale mismatch: discrete borehole point data is extrapolated to the entire domain through simple interpolation, which does not solve the problem of scale difference between point data and surface prediction, resulting in low spatial representation accuracy; the lack of coal body structure quantification: the coal body structure (such as tectonic coal) that plays a key role in gas control in low-permeability coal seams is not accurately identified and quantified, resulting in an imprecise characterization of gas occurrence patterns. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this application provides a method for accurately predicting the gas content of low-permeability coal seams based on dynamic fusion of geological constraints. The technical problem to be solved by this application is achieved through the following technical solution: A method for accurately predicting gas content in low-permeability coal seams based on dynamic fusion of geological constraints includes: Step 1: Collect multi-source geological data from the target area, including basic geological parameters, geophysical logging data, geological structure and geostress data, and coal body structure data, and construct a standardized feature set based on the multi-source geological data; Step 2: Using the measured gas content as the prediction label, extract the corresponding feature values ​​from the standardized feature set, construct a normalized sample dataset, and divide the sample dataset into a training set and a test set; Step 3: Dynamically filter and fuse the training set to obtain the optimal feature subset; the dynamic filtering and fusion includes initial screening based on the geological mechanism of gas occurrence and fine screening based on machine learning model calculation. Step 4: Train a preset machine learning prediction model using the optimal feature subset, and then use the test set to verify the trained machine learning prediction model until the preset accuracy target is reached, thus obtaining the trained machine learning prediction model. Step 5: Discretize the area to be predicted into a grid and assign four types of feature values ​​to the center point of each grid. Input the value into the trained machine learning prediction model and output the prediction result of the global gas content of the area to be predicted.

[0005] Optionally, step 1 includes: Step 1.1: Obtain the basic geological parameters of the target area and preprocess them to obtain preprocessed basic geological parameters; the basic geological parameters include coal seam burial depth, coal seam thickness, coal seam dip angle, roof and floor lithology, and lithology combination index; Step 1.2: Obtain geophysical logging data of the target area and preprocess it to obtain preprocessed geophysical logging data. The geophysical logging data includes sonic transit time, density, resistivity, natural gamma and compensated neutrons. Step 1.3: Obtain geological structure and geostress data of the target area, and preprocess them to obtain preprocessed geological structure and geostress data. The geological structure data includes curvature attributes and coherence attributes extracted based on three-dimensional seismic data, and the geostress data includes average stress and differential stress calculated based on well logging data. Step 1.4: Obtain coal body structure data of the target area and preprocess it to obtain preprocessed coal body structure data. The coal body structure data includes coal body structure type codes or classification probabilities obtained by classification model based on well logging response characteristics and borehole coring results. Step 1.5: The preprocessed basic geological parameters, preprocessed geophysical logging data, preprocessed geological structural data, preprocessed geostress data, and preprocessed coal body structure data are used as multi-source geological data, and a standardized feature set is constructed based on the multi-source geological data.

[0006] Optionally, the preprocessing of basic geological parameters in step 1.1 includes: removing outliers from burial depth, thickness, and dip angle, and normalizing the lithological combination index of the top and bottom plates to obtain preprocessed basic geological parameters. Step 1.2 involves preprocessing the geophysical logging data, including performing environmental correction and Z-score standardization on the geophysical logging data to obtain preprocessed geophysical logging data. Step 1.3 involves preprocessing the geological structure and geostress data, including performing gridded interpolation on the geological structure and geostress data to obtain preprocessed geological structure data and preprocessed geostress data. Step 1.4 involves preprocessing the coal body structure data, which includes classifying and encoding the coal body structure data to obtain preprocessed coal body structure data.

[0007] Optionally, step 2 includes: Step 2.1: Using the measured gas content as the prediction label, extract the corresponding feature values ​​from the standardized feature set to form a sample vector; Step 2.2: The sample vectors are processed using the Min-Max normalization method to eliminate dimensional differences and obtain a normalized sample dataset. Step 2.3: Divide the normalized sample dataset into a training set and a test set.

[0008] Optionally, step 3 includes: Step 3.1: Based on the gas occurrence mechanism, remove features unrelated to gas occurrence from the initial features of the training set to obtain the initial screening feature set; Step 3.2: Input the pre-screened feature set into the XGBoost model and calculate the importance score of each feature; Step 3.3: Using K-fold cross-validation, with the goal of maximizing the coefficient of determination R² of the test set, adaptively determine the feature importance threshold; Step 3.4: Select all features with importance scores higher than the feature importance threshold from the initial screening feature set to form the optimal feature subset.

[0009] Optionally, step 4 includes: Step 4.1: Input the optimal feature subset into the preset XGBoost model, and use the predicted label as the output to iteratively train the preset XGBoost model to obtain the XGBoost model after each iteration of training. Step 4.2: Use the test set to perform a comprehensive performance evaluation of the XGBoost model after each iteration of training to determine whether it meets the preset accuracy target. If not, modify the criteria for dynamically selecting and fusing the training set, and then return to step 4.1, or modify the preset parameters of the XGBoost model, and then return to step 4.1. Step 4.3: Once the XGBoost model trained in each iteration reaches the preset accuracy target, it is determined as a trained machine learning prediction model.

[0010] Optionally, step 4.2 includes: Step 4.2.1: Use the test set to perform a comprehensive performance evaluation of the XGBoost model after each iteration of training, and obtain the comprehensive performance evaluation results; Step 4.2.2: Determine whether the comprehensive performance evaluation result meets the preset accuracy target requirements; Step 4.2.3: If the comprehensive performance evaluation result does not meet the preset accuracy target requirement, modify the criteria for dynamic screening and fusion of the training set, and then return to step 4.1 or optimize the preset parameters of the XGBoost model, and then return to step 4.1.

[0011] Optionally, the parameters of the XGBoost model are optimized using a grid search method; the preset accuracy targets include: coefficient of determination R² ≥ 0.85, root mean square error RMSE ≤ 0.8 m³ / t, and mean absolute error MAE ≤ 0.5 m³ / t.

[0012] Optionally, step 5 includes: Step 5.1: Discretize the region to be predicted into a grid with a spacing of 50m×50m to generate a grid; Step 5.2: Use Kriging interpolation to assign four types of feature values ​​to the center point of each grid, thus obtaining the grid feature set; Step 5.3: Dynamically filter and fuse the set of grid features to obtain a subset of grid features; Step 5.4: Input the grid feature subset into the trained machine learning prediction model and output the global gas content prediction result of the region to be predicted.

[0013] Optionally, after step 5, the method for accurately predicting the gas content of low-permeability coal seams based on dynamic fusion of geological constraints further includes: Based on the global gas content prediction results, a gas content contour map or a three-dimensional distribution model is generated. The aforementioned gas content contour map or three-dimensional distribution model can be used to guide the layout design of gas drainage boreholes, early warning of gas outburst risks in working faces, or the delineation of dangerous areas for coal and gas outbursts.

[0014] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages: This application discloses a method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion under geological constraints. The method first constructs and standardizes a multi-source data system encompassing four major categories: basic geology, geophysical logging, geological structure and stress, and coal body structure. Second, a sample set is constructed using measured gas content as a label. An optimal feature subset is selected through a dynamic feature fusion mechanism combining initial screening based on geological mechanisms and model-driven fine screening. Subsequently, machine learning models such as XGBoost are used for training and validation. Once the required accuracy is achieved, gridded interpolation and prediction are performed on the target area to generate a gas content distribution map. This application integrates multi-source data and embeds a geological constraint mechanism, solving the problems of traditional methods such as single data source, neglect of geological origins, and poor adaptability. It enables accurate characterization of the spatial heterogeneity of gas occurrence, improves prediction accuracy, and provides a reliable basis for gas extraction design and disaster early warning. Attached Figure Description

[0015] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints provided in this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] In the relevant descriptions of this embodiment, the terms "including," "containing," and "possessing" are all open terms and are generally understood to include but not be limited to; the term "at least one" is generally understood to mean one or more, where "multiple" refers to two or more; the term "at least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items, for example, "at least one of a, b, or c", or "at least one of a, b, and c", which can all mean: a, b, c, ab (i.e., a and b), ac, bc, or abc, where a, b, and c can be single or multiple; the symbol "A / B" is used to describe the selection relationship of associated objects, generally indicating an "or" relationship.

[0019] In the following description of the embodiments, the terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms "a" and "the" as used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0020] Those skilled in the art should understand that, in the following description of the embodiments of this application, the sequence of numbers does not imply the order of execution. Some or all steps may be executed in parallel or sequentially. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0021] Those skilled in the art will understand that the numerical ranges in the embodiments of this application should be understood to specifically disclose each intermediate value between the upper and lower limits of the range. Any stated value or intermediate value within a stated range, as well as any other stated value or each smaller range between intermediate values ​​within a range, are also included within this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.

[0022] Unless otherwise stated, the technical / scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. While this application describes only preferred methods and materials, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this application. All references to this specification are incorporated by way of citation to disclose and describe the methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.

[0023] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0024] This application first constructs a multi-level, multi-source data system covering four major categories: basic geology, geophysical logging, geological structure and geostress, and coal body structure. Each type of data undergoes preprocessing such as outlier removal, environmental correction, and standardization. Then, using laboratory-measured gas content as a label, corresponding feature values ​​are matched to each borehole point to form a sample vector. After Min-Max normalization to eliminate dimensional differences, the training and test sets are divided in a 7:3 ratio. Its core innovation lies in establishing a dynamic feature fusion mechanism of "primary screening based on geological mechanisms + model-driven fine screening." First, irrelevant features are removed based on gas occurrence mechanisms. Then, the importance of features is calculated using the XGBoost model, and combined with 5-fold cross-validation, an adaptive threshold is set to select the optimal feature subset, achieving targeted fusion of multi-source features. Subsequently, the XGBoost model with optimized parameters using grid search was trained and validated. After the model achieved the performance targets of R²≥0.85, RMSE≤0.8m³ / t, and MAE≤0.5m³ / t, the prediction area was discretized into a 50m×50m grid. Kriging interpolation was used to assign feature subsets to the grid points, which were then input into the model to complete the prediction of gas content across the entire area and generate visualization results. Ultimately, this can be directly applied to engineering practices such as gas drainage borehole design, gas outburst early warning at working faces, and outburst hazard zone delineation.

[0025] like Figure 1 As shown, this application provides a method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints, including: Step 1: Collect multi-source geological data from the target area, including basic geological parameters, geophysical logging data, geological structure and geostress data, and coal body structure data, and construct a standardized feature set based on the multi-source geological data; Step 2: Using the measured gas content as the prediction label, extract the corresponding feature values ​​from the standardized feature set, construct a normalized sample dataset, and divide the sample dataset into a training set and a test set; Step 3: Dynamically filter and fuse the training set to obtain the optimal feature subset; the dynamic filtering and fusion includes initial screening based on the geological mechanism of gas occurrence and fine screening based on machine learning model calculation. This application dynamically selects the optimal feature subset that is strongly correlated with gas content from multi-source features to improve the model's generalization ability.

[0026] Step 4: Train a preset machine learning prediction model using the optimal feature subset, and then use the test set to verify the trained machine learning prediction model until the preset accuracy target is reached, thus obtaining the trained machine learning prediction model. Step 5: Discretize the area to be predicted into a grid and assign four types of feature values ​​to the center point of each grid. Input the value into the trained machine learning prediction model and output the prediction result of the global gas content of the area to be predicted.

[0027] In one specific embodiment of this application, step 1 includes: Step 1.1: Obtain the basic geological parameters of the target area and preprocess them to obtain preprocessed basic geological parameters; the basic geological parameters include coal seam burial depth (unit: m), coal seam thickness (unit: m), coal seam dip angle (unit: °), roof and floor lithology (qualitative classification) and lithology combination index (quantitative calculation: assigned a value of 1-5 according to the degree of lithological compactness, with higher values ​​for higher compactness); The preprocessing of basic geological parameters in this application includes: removing outliers from burial depth, thickness, and dip angle (using the 3σ criterion) and normalizing the lithological combination index of the top and bottom plates to obtain preprocessed basic geological parameters; Step 1.2: Obtain geophysical logging data for the target area and preprocess it to obtain preprocessed geophysical logging data. The geophysical logging data includes sonic transit time (unit: μs / m), density (unit: g / cm³), and resistivity (unit: Ω). . m), natural gamma (unit: API), compensated neutron (unit: pu); This application uses an environmental correction formula to eliminate the influence of wellbore diameter and mud density, and standardizes the dimensions using Z-Score. The formula is as follows: ; in The mean, The standard deviation is denoted as .

[0028] Step 1.3: Obtain the geological structure and geostress data of the target area, and preprocess them to obtain preprocessed geological structure and geostress data. The geological structure data includes curvature attributes and coherence attributes extracted based on 3D seismic data. The geostress data includes average stress and differential stress calculated based on well logging data. The curvature attributes and coherence attributes include maximum curvature (unit: 1 / m), minimum curvature (unit: 1 / m), Gaussian curvature (unit: 1 / m²), coherence attribute (dimensionless), and variance attribute (dimensionless).

[0029] The geological structural data in this application were extracted using seismic attribute inversion software, and a global structural parameter field was generated using Kriging interpolation. The geostress data were derived from imaging logging (wellbore collapse, induced fracture) data, or sonic and density logging data. This application calculated the formation fracture pressure gradient (unit: MPa / m) based on the Eaton method, converting it into mean stress (unit: MPa) and differential stress (unit: MPa). This application generated a global geostress scalar field data through vector interpolation, i.e., preprocessed geostress data.

[0030] Step 1.4: Obtain coal body structure data of the target area and preprocess it to obtain preprocessed coal body structure data. The coal body structure data includes coal body structure type codes or classification probabilities obtained by classification model based on well logging response characteristics and borehole coring results. This application classifies and encodes the coal body structure data to obtain preprocessed coal body structure data. Based on borehole core observation results (primary structural coal, fractured coal, granular coal, mylonite), and combined with well logging response characteristics, this application establishes classification criteria (e.g., mylonite: sonic transit time > 350 μs / m, resistivity < 50 Ω). . m, density < 1.3 g / cm³); using the pre-processed logging curves as input, a classification model is trained using the random forest algorithm (training set accounts for 70%) to classify the coal body structure of the entire well section; the classification results are encoded (primary structure coal = 1, fractured coal = 2, granular coal = 3, mylonite = 4), or the classification probability is output (dimensionless, 0-1).

[0031] Step 1.5: The preprocessed basic geological parameters, preprocessed geophysical logging data, preprocessed geological structural data, preprocessed geostress data, and preprocessed coal body structure data are used as multi-source geological data, and a standardized feature set is constructed based on the multi-source geological data.

[0032] This application can also employ Brillouin optical time-domain reflectometry (BOTDR) technology to continuously monitor gas concentration (accuracy ±1.5%) and temperature distribution along the pipeline, with a monitoring distance of up to 10 km, suitable for complex tunnel environments. Compared with traditional point sensors, its advantage is that it does not require densely arranged measuring points, reducing installation and maintenance costs; its disadvantage is lower spatial resolution and slightly weaker ability to monitor localized minute concentration changes. This scheme is suitable for macroscopic monitoring of long-distance extraction pipelines.

[0033] In one specific embodiment of this application, step 2 includes: Step 2.1: Using the measured gas content as the prediction label, extract the corresponding feature values ​​from the standardized feature set to form a sample vector; This application uses the gas content value (unit: m³ / t) measured by laboratory desorption method as the prediction label. For each borehole point with measured gas content, the feature values ​​corresponding to the four major categories of data in step 1 are extracted to form a sample vector (e.g., [burial depth = 850m, thickness = 3.2m, sonic transit time = 320μs / m, maximum curvature = 0.0021 / m, average stress = 25MPa, coal body structure code = 2]). Step 2.2: The sample vectors are processed using the Min-Max normalization method to eliminate dimensional differences and obtain a normalized sample dataset. This application uses Min-Max normalization (normalization interval [0,1]) to eliminate dimensional differences, and the formula is as follows: ; Step 2.3: Divide the normalized sample dataset into a training set and a test set.

[0034] This application randomly divides the dataset into a training set (for model training) and a test set (for model validation) in a 7:3 ratio.

[0035] In one specific embodiment of this application, step 3 includes: Step 3.1: Based on the gas occurrence mechanism, remove features unrelated to gas occurrence from the initial features of the training set to obtain the initial screening feature set; The gas occurrence mechanism in this application is based on the adsorption-desorption-transport mechanism, eliminating features unrelated to gas occurrence. For example, natural gamma is not directly related to gas occurrence and is therefore excluded.

[0036] Step 3.2: Input the initial screening feature set into the XGBoost model and calculate the importance score of each feature; the importance score of each feature in this application is based on the gain value of feature splitting.

[0037] Step 3.3: Using K-fold cross-validation, with the goal of maximizing the coefficient of determination R² of the test set, adaptively determine the feature importance threshold; This application employs 5-fold cross-validation, aiming to maximize R² on the test set, and adaptively determines the feature importance threshold. Example: the 90th percentile of feature importance on the training set.

[0038] Step 3.4: Select all features with importance scores higher than the feature importance threshold from the initial screening feature set to form the optimal feature subset.

[0039] This application selects a subset of features with scores higher than a threshold as the final model input (differentiated feature combinations are automatically output for different mining areas).

[0040] In one specific embodiment of this application, step 4 includes: Step 4.1: Input the optimal feature subset into the preset XGBoost model, and use the predicted label as the output to iteratively train the preset XGBoost model to obtain the XGBoost model after each iteration of training. This application prefers the gradient boosting decision tree model (XGBoost), and the model parameters are optimized through grid search: learning rate 0.05-0.2, tree depth 3-8, and number of leaf nodes 10-30.

[0041] This application targets unstructured data (coal seam stress waveforms, pump vibration spectra) and employs a Transformer architecture for feature extraction, combined with a Temporal Convolutional Network (TCN) to construct a prediction model. Experimental data shows that the model's prediction latency is reduced from 10s with DBN to 5s, and the response speed to sudden gas outbursts is improved by 20%. However, the model parameter size needs to be increased, requiring higher computing power, making it suitable for highly dynamic gas outburst scenarios.

[0042] Step 4.2: Use the test set to perform a comprehensive performance evaluation of the XGBoost model after each iteration of training to determine whether it meets the preset accuracy target. If not, modify the criteria for dynamically selecting and fusing the training set, and then return to step 4.1, or modify the preset parameters of the XGBoost model, and then return to step 4.1. This application uses a test set for performance evaluation. The core metrics include: coefficient of determination (R²): reflecting the goodness of fit of the model, with a target value ≥ 0.85; root mean square error (RMSE): reflecting the prediction bias, with a target value ≤ 0.8 m³ / t; mean absolute error (MAE): reflecting the average bias, with a target value ≤ 0.5 m³ / t. If the target is not met, return to step 3 to adjust the feature selection threshold or step 4.1 to optimize the model parameters.

[0043] Step 4.3: Once the XGBoost model trained in each iteration reaches the preset accuracy target, it is determined as a trained machine learning prediction model.

[0044] In one specific embodiment of this application, step 4.2 includes: Step 4.2.1: Use the test set to perform a comprehensive performance evaluation of the XGBoost model after each iteration of training, and obtain the comprehensive performance evaluation results; Step 4.2.2: Determine whether the comprehensive performance evaluation result meets the preset accuracy target requirements; Step 4.2.3: If the comprehensive performance evaluation result does not meet the preset accuracy target requirement, modify the criteria for dynamic screening and fusion of the training set, and then return to step 4.1 or optimize the preset parameters of the XGBoost model, and then return to step 4.1.

[0045] In one specific embodiment of this application, step 5 includes: Step 5.1: Discretize the region to be predicted into a grid with a spacing of 50m×50m to generate a grid; Step 5.2: Use Kriging interpolation to assign four types of feature values ​​to the center point of each grid, thus obtaining the grid feature set; Step 5.3: Dynamically filter and fuse the set of grid features to obtain a subset of grid features; Step 5.4: Input the grid feature subset into the trained machine learning prediction model and output the global gas content prediction result of the region to be predicted.

[0046] This application discretizes the prediction area into a 50m×50m grid to generate grid center points; feature assignment: spatial interpolation (Kriging method) is used to assign four major categories of feature values ​​to each grid center point, and a feature subset is obtained after filtering in step 3; prediction calculation: the grid feature subset is input into the trained model, and the predicted gas content value of each grid is output.

[0047] In one specific embodiment of this application, after step 5, the method for accurately predicting the gas content of low-permeability coal seams based on dynamic fusion of geological constraints further includes: Based on the global gas content prediction results, a gas content contour map or a three-dimensional distribution model is generated. The aforementioned gas content contour map or three-dimensional distribution model can be used to guide the layout design of gas drainage boreholes, early warning of gas outburst risks in working faces, or the delineation of dangerous areas for coal and gas outbursts.

[0048] This application generates a planar contour map of gas content (contour interval 0.5 m³ / t) or a 3D visualization model. In engineering applications, the global gas content prediction results are applied to: gas drainage borehole design: denser borehole layout in high gas content areas (>15 m³ / t), and optimized borehole spacing in low gas content areas (<8 m³ / t); gas outburst early warning at working faces: calculating the gas outburst volume at working faces based on the predicted values ​​and setting graded early warning thresholds; outburst hazard zone delineation: combining gas content and coal body structure data to delineate outburst hazard zones, threat zones, and safe zones.

[0049] This application can delineate areas with gas content exceeding a critical value (e.g., 8 m³ / t) as gas-enriched target zones based on the overall gas content prediction results. This application employs three-dimensional drilling trajectory design software, based on the principles of "penetrating the maximum gas enrichment volume" and "avoiding geological risk areas (such as faults)," to intelligently generate wellbore trajectories for directional long boreholes or horizontal wells drilled from the surface. Directional drilling technology is used, and the wellbore trajectory is strictly followed to ensure that the borehole accurately penetrates one or more gas-enriched target zones. In the target borehole, one or more of the following waterless permeability enhancement measures are sequentially or selectively implemented: (1) High-pressure air blasting for enhanced permeability: The high-pressure air blaster is lowered into the gas-rich section of the borehole, and the high-pressure gas shock wave generated instantaneously causes the coal body to generate and expand a multi-level fracture network, thereby achieving waterless permeability enhancement.

[0050] (2) Physical vibration wave excitation permeation enhancement: A high-power sound wave / vibration wave generator is installed at the orifice or inside the orifice to emit mechanical waves of a specific frequency to the coal seam, using the wave energy to promote gas desorption and improve fracture connectivity.

[0051] (3) Controllable microwave radiation-induced permeation enhancement: The microwave radiation antenna is sent into the enriched section of the borehole. Through the selective heating effect of microwaves on the coal body, the temperature of the coal body is greatly increased, the gas desorption dynamics process is accelerated, and the gas dissipation rate is increased.

[0052] (4) Intelligent sampling and dynamic optimization based on the Internet of Things.

[0053] The following steps are used to achieve intelligent sampling and dynamic optimization based on the Internet of Things: 1. Install an intelligent extraction device at the directional borehole opening. This device integrates an electric control valve, a gas flow / concentration / pressure sensor, and a data acquisition and transmission unit.

[0054] 2. Establish a real-time feedback optimization system based on extraction data. Utilize IoT technology to collect real-time data on extraction negative pressure, pure gas flow rate, and concentration from each borehole.

[0055] 3. Construct an extraction effect evaluation model. When the pure gas flow rate of a borehole is continuously lower than the set threshold, the system will automatically or remotely command the engineer to start the permeation enhancement measures in the controllable microwave radiation excitation permeation enhancement for secondary excitation; or automatically adjust the extraction negative pressure to the optimal range to achieve intelligent dynamic management of one policy per borehole.

[0056] This application utilizes a high-efficiency permanent magnet motor-driven high-power extraction pump (efficiency increased to 90%, power range 100-500kW), paired with an electro-hydraulic servo valve (response time <100ms, control accuracy ±0.1%), which can shorten the control cycle to 15s. This solution is suitable for the emergency extraction needs of high-risk gas outburst mines, and can rapidly increase extraction capacity after the appearance of gas outburst signs, reducing the risk of outbursts and excessive gas levels. The disadvantages are higher equipment investment costs (30% higher than conventional solutions) and more stringent maintenance requirements for the hydraulic system.

[0057] This application can overcome the shortcomings of traditional prediction methods, such as single data source, neglect of geological constraints, and poor adaptability. It improves prediction accuracy by 20%-30% and reduces RMSE by more than 35% compared with traditional methods. It can accurately characterize the spatial heterogeneity of gas occurrence in low-permeability coal seams and can automatically optimize feature combinations according to different geological conditions of mining areas. At the same time, it can reduce gas control costs by 15%-20% and can be transferred to other mineral resource prediction fields. It has the advantages of high accuracy, adaptability, engineering practicality and technical scalability.

[0058] This application utilizes multi-source data fusion and an intelligent evaluation model, enabling the system to identify high-concentration areas in real time and prioritize their extraction. Through dynamic evaluation and precise control, the extraction time to meet standards in high-concentration areas is reduced by 50%, the system's average extraction efficiency is increased to 82%, and the number of gas exceedances is reduced by 75%. The variable frequency pump unit in this application, combined with an intelligent start-stop strategy, reduces energy consumption by 28%; the utilization rate of low-concentration gas enrichment reaches 90%, resulting in an annual CO2 reduction equivalent to approximately 12,000 tons (based on an annual gas extraction of 5 million m³). 3 (Calculation). This application employs a data fusion and global optimization mechanism to achieve safety indicators (gas concentration <1%), energy efficiency indicators (energy consumption per ton of coal extraction <1.5kWh), and emission reduction indicators (carbon emissions per unit extraction <0.8kgCO2 / m³). 3 This application achieves simultaneous optimization of gas extraction, meeting the dual goals of intelligent and green mining in coal mines. It realizes the transformation of gas extraction from manual experience-driven to data-driven intelligence, and simultaneously constructs an integrated green mining model of "extraction-utilization-emission reduction" through gas resource utilization and heat recovery, demonstrating significant socio-economic benefits.

[0059] To illustrate the technical effects of this application, a low-permeability coal mining area is selected as the implementation object. This mining area has a coal seam permeability of 0.01-0.1 mD, a coal-bearing stratum thickness of 50-80 m, multiple faults and folds, widespread tectonic coal distribution, and measured gas content of 8-22 m³ / h. 3 / t. 5.2 Implementation Steps Details 5.2.1 Data Acquisition and Preprocessing Basic Geological Parameters: Collect burial depth (750-1050m), thickness (2.1-4.5m), and dip angle (5-18°) of 30 boreholes. The lithology of the top and bottom plates is mainly sandstone and mudstone, with a combination index of 2-4; Well Logging Data: Collect sonic transit time (280-420μs / m) and density (1.2-1.5g / cm³) of 25 wells. 3 Five types of curves, including , were standardized after environmental correction; tectonic-geostress data: curvature, coherence, and other attributes were extracted from the 3D seismic data volume interpretation, and the average stress (20-35MPa) was calculated using the Eaton method; coal structure identification: coal structure classification based on 10 core wells (30% primary structure coal, 45% fractured coal, 25% granular / mylonite coal), a random forest classification model was trained, with an identification accuracy of 91%. 5.2.2 Sample dataset and feature engineering Sample size: 120 measured gas content samples (84 in the training set, 36 in the test set); Number of initial features: 23 (4 in basic geology, 5 in well logging, 10 in tectonic-geostress, and 4 in coal structure); Preprocessing: after Min-Max normalization, the feature values ​​are distributed in the [0,1] interval. 5.2.3 Dynamic Feature Screening and Geological Mechanism Initial Screening: Two irrelevant features, natural gamma and coal seam dip angle, were removed, leaving 21 features; Model-driven Fine Screening: XGBoost was used to calculate feature importance and select 12 high-contribution features (such as burial depth, sonic transit time, maximum curvature, average stress, coal body structure coding, etc.); Threshold Setting: Based on 5-fold cross-validation, the feature importance threshold was determined to be 0.035.

[0060] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. Each embodiment focuses on describing the differences from other embodiments.

[0061] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of this application.

Claims

1. A method for accurately predicting the gas content of low-permeability coal seams based on dynamic fusion of geological constraints, characterized in that, include: Step 1: Collect multi-source geological data from the target area, including basic geological parameters, geophysical logging data, geological structure and geostress data, and coal body structure data, and construct a standardized feature set based on the multi-source geological data; Step 2: Using the measured gas content as the prediction label, extract the corresponding feature values ​​from the standardized feature set, construct a normalized sample dataset, and divide the sample dataset into a training set and a test set; Step 3: Dynamically filter and fuse the training set to obtain the optimal feature subset; The dynamic screening and fusion includes preliminary screening based on the geological mechanism of gas occurrence, and fine screening based on machine learning model calculation. Step 4: Train a preset machine learning prediction model using the optimal feature subset, and then use the test set to verify the trained machine learning prediction model until the preset accuracy target is reached, thus obtaining the trained machine learning prediction model. Step 5: Discretize the area to be predicted into a grid and assign four types of feature values ​​to the center point of each grid. Input the value into the trained machine learning prediction model and output the prediction result of the global gas content of the area to be predicted.

2. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 1, characterized in that, Step 1 includes: Step 1.1: Obtain the basic geological parameters of the target area and preprocess them to obtain preprocessed basic geological parameters; the basic geological parameters include coal seam burial depth, coal seam thickness, coal seam dip angle, roof and floor lithology, and lithology combination index; Step 1.2: Obtain geophysical logging data of the target area and preprocess it to obtain preprocessed geophysical logging data. The geophysical logging data includes sonic transit time, density, resistivity, natural gamma and compensated neutrons. Step 1.3: Obtain geological structure and geostress data of the target area, and preprocess them to obtain preprocessed geological structure and geostress data. The geological structure data includes curvature attributes and coherence attributes extracted based on three-dimensional seismic data, and the geostress data includes average stress and differential stress calculated based on well logging data. Step 1.4: Obtain coal body structure data of the target area and preprocess it to obtain preprocessed coal body structure data. The coal body structure data includes coal body structure type codes or classification probabilities obtained by classification model based on well logging response characteristics and borehole coring results. Step 1.5: The preprocessed basic geological parameters, preprocessed geophysical logging data, preprocessed geological structural data, preprocessed geostress data, and preprocessed coal body structure data are used as multi-source geological data, and a standardized feature set is constructed based on the multi-source geological data.

3. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 2, characterized in that, Step 1.1 involves preprocessing the basic geological parameters, including removing outliers from the burial depth, thickness, and dip angle, and normalizing the lithological combination index of the top and bottom plates to obtain the preprocessed basic geological parameters. Step 1.2 involves preprocessing the geophysical logging data, including performing environmental correction and Z-score standardization on the geophysical logging data to obtain preprocessed geophysical logging data. Step 1.3 involves preprocessing the geological structure and geostress data, including performing gridded interpolation on the geological structure and geostress data to obtain preprocessed geological structure data and preprocessed geostress data. Step 1.4 involves preprocessing the coal body structure data, which includes classifying and encoding the coal body structure data to obtain preprocessed coal body structure data.

4. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 1, characterized in that, Step 2 includes: Step 2.1: Using the measured gas content as the prediction label, extract the corresponding feature values ​​from the standardized feature set to form a sample vector; Step 2.2: The sample vectors are processed using the Min-Max normalization method to eliminate dimensional differences and obtain a normalized sample dataset. Step 2.3: Divide the normalized sample dataset into a training set and a test set.

5. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 1, characterized in that, Step 3 includes: Step 3.1: Based on the gas occurrence mechanism, remove features unrelated to gas occurrence from the initial features of the training set to obtain the initial screening feature set; Step 3.2: Input the pre-screened feature set into the XGBoost model and calculate the importance score of each feature; Step 3.3: Using K-fold cross-validation, with the goal of maximizing the coefficient of determination R² of the test set, adaptively determine the feature importance threshold; Step 3.4: Select all features with importance scores higher than the feature importance threshold from the initial screening feature set to form the optimal feature subset.

6. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 1, characterized in that, Step 4 includes: Step 4.1: Input the optimal feature subset into the preset XGBoost model, and use the predicted label as the output to iteratively train the preset XGBoost model to obtain the XGBoost model after each iteration of training. Step 4.2: Use the test set to perform a comprehensive performance evaluation of the XGBoost model after each iteration of training to determine whether it meets the preset accuracy target. If not, modify the criteria for dynamically selecting and fusing the training set, and then return to step 4.1, or modify the preset parameters of the XGBoost model, and then return to step 4.

1. Step 4.3: Once the XGBoost model trained in each iteration reaches the preset accuracy target, it is determined as a trained machine learning prediction model.

7. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 6, characterized in that, Step 4.2 includes: Step 4.2.1: Use the test set to perform a comprehensive performance evaluation of the XGBoost model after each iteration of training, and obtain the comprehensive performance evaluation results; Step 4.2.2: Determine whether the comprehensive performance evaluation result meets the preset accuracy target requirements; Step 4.2.3: If the comprehensive performance evaluation result does not meet the preset accuracy target requirement, modify the criteria for dynamic screening and fusion of the training set, and then return to step 4.1 or optimize the preset parameters of the XGBoost model, and then return to step 4.

1.

8. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 7, characterized in that, The parameters of the XGBoost model are optimized using a grid search method; The preset accuracy targets include: coefficient of determination R² ≥ 0.85, root mean square error RMSE ≤ 0.8 m³ / t, and mean absolute error MAE ≤ 0.5 m³ / t.

9. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 1, characterized in that, Step 5 includes: Step 5.1: Discretize the region to be predicted into a grid with a spacing of 50m×50m to generate a grid; Step 5.2: Use Kriging interpolation to assign four types of feature values ​​to the center point of each grid, thus obtaining the grid feature set; Step 5.3: Dynamically filter and fuse the set of grid features to obtain a subset of grid features; Step 5.4: Input the grid feature subset into the trained machine learning prediction model and output the global gas content prediction result of the region to be predicted.

10. The method for accurate prediction of gas content in low-permeability coal seams based on dynamic fusion of geological constraints according to claim 1, characterized in that, Following step 5, the method for accurately predicting the gas content of low-permeability coal seams based on dynamic fusion of geological constraints further includes: Based on the global gas content prediction results, a gas content contour map or a three-dimensional distribution model is generated. The aforementioned gas content contour map or three-dimensional distribution model can be used to guide the layout design of gas drainage boreholes, early warning of gas outburst risks in working faces, or the delineation of dangerous areas for coal and gas outbursts.

Images