Coal seam gas pressure in-situ measurement method while drilling based on LSTM (Long Short Term Memory)

Abstract

The invention relates to a coal seam gas pressure in-situ measurement method while drilling based on LSTM (Long Short Term Memory), and belongs to the technical field of coal mine gas prevention and control. Aiming at the problems of long pressure measurement period, high hole sealing requirement, large interference of geological conditions, low automation degree and the like in the existing gas pressure measurement technology, the method comprises the following steps of: establishing a gas emission quantity theoretical calculation model, combining mining area coal seam geology and gas parameters, and theoretically calculating to generate an emission quantity time sequence data set covering different gas pressures; collecting field actual data, aligning the field actual data with a data set, and constructing a dual-channel LSTM neural network model fusing time sequence dynamic characteristics and static prior knowledge for training; and finally, embedding the trained model into an anti-explosion edge calculation terminal to realize in-situ rapid inversion of the gas pressure while drilling. According to the invention, the rapid, accurate and automatic measurement of the gas pressure is realized, the interference to the mining operation is obviously reduced, and the efficiency of coal mine gas disaster prevention and control is improved.

Description

Technical Field

[0001] This invention belongs to the field of coal mine gas control technology, and relates to an in-situ measurement method for coal seam gas pressure based on LSTM. Background Technology

[0002] Accurate measurement of coal seam gas pressure is a crucial step in identifying coal and gas outburst hazards and assessing coal seam gas resources. However, influenced by various factors such as geological structure, burial depth, and variations in coal seam thickness, the occurrence of coal seam gas exhibits significant heterogeneity, and the distribution of coal seam outburst hazards also shows regional heterogeneity. This greatly increases the difficulty and complexity of coal mine gas disaster management.

[0003] Currently, underground gas pressure measurement technology in coal mines is mainly divided into two categories: direct measurement and indirect measurement. Direct measurement can be further divided into passive and active measurement. Passive measurement involves drilling a borehole in the coal seam, installing a pressure measuring pipe, and sealing the borehole. The pressure is monitored by a pressure gauge, and the gas pressure is obtained once the pressure value stabilizes. This method is relatively mature and low-cost, but it suffers from problems such as difficulty in ensuring the quality of the sealing, leading to gas leakage, and excessively long measurement cycles, which seriously affect the continuous operation of coal mining and excavation. Active measurement involves injecting gas at a certain pressure into the pressure measuring chamber of the borehole to accelerate the pressure balance process. This method can shorten the measurement time to some extent and overcome some of the effects of sealing leakage, but it is technically complex, expensive, and susceptible to disturbances during borehole construction and the influence of roadway pressure relief. Indirect measurement typically uses historical logging data, coal seam gas content, or gas emission rates, and calculates the gas pressure value using the Langmuir equation. Due to the generally lower accuracy and reliability of indirect measurement, direct measurement is still preferred in most important situations.

[0004] As the coal mining industry moves towards intensive production, rapid mining, and deeper coal seams, traditional gas pressure measurement technologies have revealed significant limitations. Severe lag in measurement cycles, uncertainty in measurement accuracy, and low levels of automation have become prominent issues hindering efficient and safe coal mine production. In recent years, machine learning-based methods have made some progress in coal seam gas parameter prediction research, but these studies mainly focus on predictive modeling of directly observable, continuously changing time-series data such as gas concentration or gas emission. However, gas pressure itself is a steady-state parameter that cannot be directly observed, lacking sufficient real-time monitoring data, making it difficult to directly use supervised learning methods for modeling and inversion. Long Short-Term Memory (LSTM) networks, through their unique gating mechanisms (input gate, forget gate, and output gate), can effectively capture long-term dependencies in time series, automatically learning and remembering key historical information. Furthermore, LSTM can learn complex nonlinear mapping relationships from limited labeled data; its ability to process hundreds or even thousands of time-step sequence data enables it to extract pressure-sensitive features from the gas emission time series in the early stages of borehole exposure, thus achieving high-precision inversion.

[0005] Therefore, there is an urgent need for a new method that can overcome the above-mentioned defects. By combining LSTM time series modeling with gas emission dynamics, the intrinsic correlation between gas emission and gas pressure can be constructed, enabling rapid and accurate in-situ inversion of gas pressure. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide an in-situ measurement method for coal seam gas pressure during drilling based on LSTM.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for in-situ measurement of coal seam gas pressure during drilling based on LSTM includes the following steps: S1: Establish a theoretical calculation model for gas emission based on gas pressure; S2: Obtain the actual historical data information of the current coal seam in the mining area, and preprocess the data information; S3: Using the historical coal seam information data obtained in S2, perform theoretical calculations on the time-series gas emission rate, generate the corresponding time-series gas emission rate curve, and form a structured dataset. S4: Collect on-site coal seam information data and actual gas emission time-series data, perform preprocessing, and add the structured dataset obtained in S3. Gaussian noise was removed and aligned with the actual field measurement data. S5: Construct a dual-channel LSTM gas pressure inversion neural network model, and train and validate the model; S6: Embed the trained dual-channel LSTM gas pressure inversion neural network model into the explosion-proof edge computing terminal.

[0008] Furthermore, in S1, the theoretical calculation model for gas emission based on gas pressure is expressed as follows:

[0009] in, This refers to the time-series gas emission rate. P The gas pressure of the coal seam to be measured. The porosity of the coal seam. k For coal seam permeability, a and b Let be the adsorption constant of coal. s The effective stress coefficient of coal, h For coal seam thickness, t For time, x This is the borehole location information.

[0010] Furthermore, in S3, the gas emission time-series curve is calculated using the following formula:

[0011] in, This refers to the amount of gas emitted from the borehole wall during the drilling process. The specific formula for calculating the gas emission rate from the borehole wall in the stopped drilling section is as follows:

[0012]

[0013] in, and These represent the drilling duration and the shutdown duration, respectively, in seconds (s). The instantaneous gas outburst intensity caused by coal breaking, expressed in meters (m). 3 / (m 2 ·s); and These are the geometric domains of the hole's peripheral wall and bottom wall, respectively. The Darcy velocity on the wall is expressed in m / s. k Coal permeability, in meters (m). 2 ; The viscosity is the fluid dynamic viscosity, expressed in Pa·s. The gas pressure gradient on the wall surface; The radius of the borehole is in meters. This refers to the drilling speed, expressed in m / min. The length of the borehole is in meters (m).

[0014] Furthermore, in S2, the actual historical data information includes geological structure data, coal sample porosity, permeability, adsorption constant, historical gas pressure and its changing trend, and the effective stress coefficient of coal.

[0015] Furthermore, in S3, by traversing different gas pressures... Calculate and generate the corresponding time-series curve of gas emission. To form a structured dataset .

[0016] Furthermore, in S4, the preprocessing includes data cleaning, handling missing values ​​and outliers.

[0017] Furthermore, in S5, the main input variable of the dual-channel LSTM gas pressure inversion neural network model is the time series data of gas emission. ,in T The number of sampling points; the auxiliary input variable is coal seam information data, including porosity. Penetration rate k Adsorption constant a and b Effective stress coefficient and coal seam thickness h and location information x Normalization is performed; the output variable is the coal seam gas pressure.

[0018] Furthermore, the dual-channel LSTM gas pressure inversion neural network model includes: The temporal feature extraction branch employs a two-layer bidirectional LSTM with 128 hidden units per layer, and the input is a gas outflow sequence. The output is a 256-dimensional hidden state vector; The prior knowledge encoding branch uses a two-layer fully connected network, with normalized coal seam parameters as input. , k , a , b , s , h The output is a 16-dimensional encoded vector; The feature concatenation layer concatenates the hidden state vector and the encoded vector, and then outputs the predicted gas pressure value through a three-layer decoder. .

[0019] Furthermore, in S5, the loss function used for model training is:

[0020] in, Mean square error; Physical constraints; For boundary constraints; These are the weighting coefficients. This represents the actual stress value. To predict pressure values ​​for the model.

[0021] Furthermore, in S5, during model training, the training dataset includes the structured dataset obtained from theoretical calculations in S3 and the field test data. The structured dataset is divided into training set, validation set, and test set in a 7:2:1 ratio, while the field test data is divided into training set, validation set, and test set in a 4:4:2 ratio.

[0022] The beneficial effects of this invention are as follows: The beneficial effects of this invention are reflected in the following aspects: (1) This invention breaks through the technical bottleneck of traditional gas pressure measurement, which must rely on long-term sealing and stabilization, and realizes a paradigm shift from passively waiting for pressure balance to actively and dynamically inverting. By utilizing the time-series modeling capability of long short-term memory networks to extract the dynamic characteristics of gas outbursts, pressure measurement can be completed quickly without additional sealing operations, significantly shortening the measurement time, greatly reducing interference with normal mining operations, and improving the efficiency of coal seam parameter acquisition.

[0023] (2) This invention proposes a dual-channel feature processing mechanism that embeds coal and rock physical parameters as prior knowledge into a neural network, which effectively solves the problem of model misjudgment caused by differences in geological conditions. By separating the processing of time-series dynamic features and static physical background, this method enables the model to accurately distinguish between changes in outburst caused by pressure changes and changes in coal quality itself, significantly enhancing the reliability, accuracy and engineering applicability of the inversion results.

[0024] (3) This invention employs a hierarchical hybrid data strategy to construct the training dataset, fully leveraging the advantages of broad coverage of theoretical calculation data and reliable on-site measured data. This strategy ensures that the model can learn general laws from rich simulation data while also guaranteeing its generalization ability and robustness in real industrial scenarios.

[0025] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0027] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0028] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0029] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0030] Please see Figure 1 This includes the following steps: S1. Establish a theoretical calculation method for gas emission based on gas pressure; S2: Obtain the actual historical data of the current coal seam in the mining area and preprocess the obtained data; S3: Using the collected historical information data of coal seams, theoretical calculations are performed on the gas emission time series to generate the corresponding gas emission time series curve, forming a structured dataset; S4: Collect and preprocess on-site coal seam information data and actual gas emission time series data, add ±5% Gaussian noise to the structured dataset in S3 and align it with the on-site measured data; S5: Construct a dual-channel LSTM gas pressure inversion neural network model, and train and validate the model; S6: Embed the trained model into the explosion-proof edge computing terminal.

[0031] In S1, the theoretical calculation model for gas emission based on gas pressure is expressed as follows:

[0032] In the formula: This refers to the time-series gas emission rate. P The gas pressure of the coal seam to be measured. The porosity of the coal seam. k Let be the coal seam permeability, and a and b be the adsorption constants of the coal. The effective stress coefficient of coal, x This is the borehole location information.

[0033] In S2, the actual historical data information of the current coal seam in the mining area is obtained, including but not limited to: obtaining geological structure data through coal mine exploration information; obtaining coal sample porosity, permeability, and adsorption constant through laboratory measurements; obtaining gas pressure and its changing trend at each measuring point through the historical monitoring system; and obtaining the effective stress coefficient of coal through the stress monitoring system.

[0034] In S3, the coal seam information collected in S2 is used to iterate through different gas pressures. Calculate and generate the corresponding time-series monitoring curve of gas emission. , , ...and ultimately form a structured dataset. .

[0035]

[0036]

[0037]

[0038] in, This refers to the amount of gas emitted from the borehole wall during the drilling process. This refers to the amount of gas emitted from the borehole wall during the drilling stoppage section. and These represent the drilling duration and the shutdown duration, respectively, in seconds. The instantaneous gas outburst intensity caused by coal breaking at the borehole wall, m 3 / (m 2 ·s); and These are the geometric domains of the hole's peripheral wall and bottom wall, respectively. Let be the Darcy velocity on the wall, in m / s; For coal permeability, m 2 ; The viscosity is the fluid dynamic viscosity, Pa·s; The gas pressure gradient on the wall surface; Let be the borehole radius, in meters. The drilling speed is expressed in m / min. The length of the borehole is in meters (m).

[0039] In S4, continuous monitoring equipment (such as infrared gas sensors and flow meters) is activated immediately after drilling to monitor and collect gas emission; the collected data is cleaned and missing and outlier values ​​are processed.

[0040] In S5, the main input variable in the LSTM gas pressure inversion neural network model is... ,in, T The number of sampling points (e.g., sampling once per second for 30 minutes). T =1800); the auxiliary input variable is the layer information data collected in S2, and the location information is also included. x Perform normalization; the output variable is the in-situ coal seam gas pressure, which is a scalar value.

[0041] Normalization method:

[0042] In S5, the model's neural network architecture uses feature concatenation to combine temporal features and prior knowledge, where the temporal feature is the gas emission rate. The prior knowledge feature is coal seam information. The two feature streams are concatenated and fused to output the final gas pressure estimate. The model uses mean squared error as the loss function.

[0043] Temporal feature extraction branch: A two-layer bidirectional LSTM is used, with 128 hidden units in each layer. The input is a gas outflow sequence of length T. The output is a 256-dimensional hidden state vector; the prior knowledge encoding branch uses a 2-layer fully connected network (32→16 units), and the input is the normalized coal seam parameters. The output is a 16-dimensional encoded vector; the two feature streams are concatenated and then passed through a 3-layer decoder (64→32→1 unit) to output the predicted pressure. .

[0044] The model loss function is:

[0045] in Mean square error, For physical constraints, For boundary constraints, These are the weighting coefficients.

[0046] in True stress level To predict pressure values ​​for the model.

[0047] In S5, the model training dataset includes the structured dataset obtained from theoretical calculations in S3 and the field-tested data. The structured dataset is divided into training, validation, and test sets in a ratio of 7:2:1, and the field-tested data is divided into training, validation, and test sets in a ratio of 4:4:2.

[0048] Example 1: Basic Implementation Process This invention includes the following steps: S101. Establish a theoretical calculation model for gas emission rate based on gas pressure. This model forms the physical foundation for subsequent numerical simulations and dataset construction. Its core lies in accurately describing the dynamic process of gas emission rate evolution over time under specific coal seam geological parameters and drilling process parameters through multiphysics coupling equations. This model can be abstractly represented by the following functional relationship:

[0049] in, express t The amount of gas emitted at any given moment. P For the coal seam gas pressure to be inverted, The porosity of the coal seam. k For coal seam permeability, a and b The Langmuir adsorption constant for coal is... The effective stress coefficient of coal, h For coal seam thickness, t For time, x This provides the borehole location information. The specific implementation of the function f is based on Darcy's law for gas flow, the mass conservation equation, and the adsorption-desorption kinetics of coal. Its detailed expression is described later.

[0050] S102. Obtain actual historical data for the target mining area. Systematically collect typical parameter ranges of the target coal seam through geological exploration reports, laboratory testing and analysis of coal core samples, and mine historical monitoring databases. These parameters constitute the prior knowledge input for the model, including but not limited to: geological structural features and coal sample porosity. Penetration rate k Adsorption constant a and b Effective stress coefficient derived from geostress field data Coal seam thickness h and historical gas pressure values P and its distribution range This multi-source heterogeneous data is cleaned, integrated, and standardized preprocessed to form a complete and consistent prior knowledge dataset.

[0051] S103. Generate a structured simulation dataset based on the theoretical model and historical data. Utilize the parameter range obtained in step S102, especially the possible range of gas pressure values. Large-scale numerical simulations were performed on a high-performance computer. For each set gas pressure value... P i By combining other fixed coal seam parameters and solving the following theoretical model, a complete, high-time-resolution theoretical gas emission time-series curve is calculated. The detailed mathematical expression of this model is as follows: Total gas emission from borehole Outflow from the drilling section and the outflow of drilling section constitute:

[0052] in, The calculation involves two cases: when =0 (initial instant): when (During drilling)

[0053] It takes effect after drilling stops. ):

[0054] The parameters in the formula are defined as follows: and These are the drilling duration and the stoppage duration, respectively, in seconds (s); The instantaneous gas outburst intensity caused by coal breaking, expressed in meters (m). 3 / (m 2 ·s); and These are the geometric domains of the hole's peripheral wall and bottom wall, respectively. The Darcy velocity on the wall is expressed in m / s. k Coal permeability, in meters (m). 2 ; The viscosity is the fluid dynamic viscosity, and the unit is Pascal-second (Pa·s). The gas pressure gradient on the wall surface; The radius of the borehole is in meters. This refers to the drilling speed, expressed in m / min. The length of the borehole is in meters (m).

[0055] Finally, all simulation results are paired and stored to form a structured dataset. Each sample contains a time-series curve of outflow and its corresponding gas pressure value as a label.

[0056] S104. Collect and preprocess on-site measured data. During downhole drilling, deploy high-frequency infrared gas sensors or precision flow meters while drilling to continuously monitor and record the actual gas emission at the borehole opening, forming measured time-series data. Simultaneously, record the coal seam parameters at the borehole location. , k , a , b , s , h ] and location x The measured data undergoes preprocessing, including denoising and outlier removal. To enhance the robustness of the simulated data and reduce the distributional discrepancy between the simulated data and the real data, the theoretical dataset generated in step S103 is preprocessed. Add Gaussian white noise with an amplitude of ±5% and align it with the measured data in terms of time scale and magnitude.

[0057] S105. Construct, train, and validate a dual-channel long short-term memory (LSTM) gas pressure inversion neural network model. The model is designed with a dual-channel input architecture.

[0058] 1. Temporal Feature Extraction Branch (Main Channel): Input is length T (For example T =1800 (corresponding to a sampling rate of 30 minutes per second) Gas emission sequence The sequence is processed by two layers of bidirectional LSTM network (128 hidden units per layer), and finally outputs a 256-dimensional feature vector to capture deep temporal patterns in the dynamic sequence of outflows.

[0059] 2. Prior knowledge encoding branch (auxiliary channel): The input is a normalized coal seam static parameter vector. , k , a , b , s , h The vector is encoded through a two-layer fully connected network (32 units in the first layer and 16 units in the second layer), outputting a 16-dimensional feature vector representing the geological background information of the coal seam.

[0060] 3. Feature Fusion and Decoding: The 256-dimensional temporal feature vector is concatenated with the 16-dimensional prior knowledge encoding vector to form a 272-dimensional fused feature vector. This fused feature is then input into a decoder consisting of three fully connected layers (64, 32, and 1 nodes respectively), ultimately outputting a scalar value, namely the predicted gas pressure. .

[0061] Model training uses a composite loss function L To simultaneously optimize prediction accuracy and introduce physical constraints: in, For mean square error loss, This represents the actual stress value. These are the model's predicted values.

[0062] The physical constraint loss ensures that the partial derivative of the predicted pressure with respect to the initial outflow is non-negative, which conforms to physical laws.

[0063] For boundary constraint losses, the penalty exceeds a reasonable range. The predicted value.

[0064] These are weighting coefficients used to balance various losses.

[0065] During training, the theoretically calculated structured dataset was divided into training, validation, and test sets in a 7:2:1 ratio, while the field-tested data was divided in a 4:4:2 ratio. Iterative training was performed using the Adam optimization algorithm, and performance was monitored on the validation set to prevent overfitting until the model converged.

[0066] S106. Model Deployment and Application. The trained and validated model parameters are solidified and embedded into an edge computing terminal with mining explosion-proof certification. This terminal is installed near the drilling rig or in a safety chamber. In practical applications, time-series data of gas emission monitored during drilling are used. Q ( t ) and the current coal seam parameters of the borehole [ , k , a , b , s , h The data is input into the terminal in real time, and the trained dual-channel LSTM model can quickly infer and output the in-situ gas pressure estimate of the currently encountered coal seam. This enables rapid and accurate in-situ measurements while drilling.

[0067] Example 2: Enhanced Data Acquisition and Processing for Complex Geological Conditions This embodiment focuses on expanding the data acquisition and processing stages, and is applicable to mining areas with complex geological conditions and large variations in coal quality.

[0068] First, given the complex geological conditions, data acquisition needs to be more refined. Geological structural data is not limited to exploration reports; it also needs to be combined with geophysical methods such as downhole channel seismic exploration and radio wave tunnel imaging to finely characterize the distribution of microstructures such as small faults and folds. Coal sample testing requires systematic sampling from different coal seams to obtain porosity data. Penetration rate k The statistical distribution characteristics of parameters (such as mean and variance) should be considered, rather than a single average. Historical gas pressure data need to be analyzed by geological unit (such as fault block or anticline limb) to determine the specific characteristics of each region. and .

[0069] Secondly, when generating the simulation dataset, Monte Carlo random sampling is used instead of simple parameter traversal. That is, parameter combinations are randomly selected within the statistical distribution range of each parameter for numerical simulation. For example, the permeability k is no longer a fixed value, but is randomly selected from a log-normal distribution that conforms to the statistical laws of the mining area. This results in a structured dataset. It better reflects the uncertainty and coupling effect of geological parameters, significantly increases sample diversity, and is closer to reality.

[0070] Finally, in the field data preprocessing stage, considering that sensor signals are more susceptible to drilling rig vibration and electromagnetic interference under complex conditions, a combination of wavelet transform and Kalman filtering is used for data cleaning, which more effectively separates the true dynamic characteristics of gas outburst from noisy signals. This embodiment effectively improves the generalization ability and predictive stability of the finally trained pressure inversion model under complex geological conditions by enhancing the diversity and quality of data sources and introducing a data generation strategy that better reflects actual geological uncertainties.

[0071] Example 3: Variations of Neural Network Models and Training Strategies This embodiment specifies and expands the neural network model and its training method.

[0072] The model structure can be appropriately adjusted according to the actual application scenario and data characteristics. For example, for applications with higher sampling frequencies and longer sequence lengths... T For larger outflow data, the number of LSTM layers can be increased to three to capture longer-term temporal dependencies. Alternatively, a more computationally efficient gated recurrent unit (GRU) can be used instead of LSTM to reduce the computational load on edge computing terminals while maintaining performance. In the prior knowledge encoding branch, an attention mechanism module can be introduced to allow the model to automatically learn different coal seam parameters (such as...). ,k , a The importance of stress inversion (etc.) is dynamically adjusted in the final feature representation, thereby more intelligently integrating prior knowledge.

[0073] In terms of training strategy, an advanced phased training method is adopted. First, the model is pre-trained using a large-scale, noisy, theoretically calculated structured dataset to allow it to initially grasp the basic physical mapping relationship between gas outburst dynamics and pressure. Then, a limited but reliable amount of field-measured data is used to fine-tune the pre-trained model, allowing the model parameters to adaptively adjust to the data distribution under real-world conditions. This pre-training-fine-tuning paradigm helps alleviate the problems of overfitting or weak generalization ability caused by insufficient field-measured data.

[0074] For composite loss functions, the constraint terms can be further refined. Physical constraint loss. This can be extended to constrain not only pressure on the initial outflow. The monotonicity of the Darcy velocity term can also be introduced. Soft constraints on correlations. Boundary constraint loss. In and The parameters can be dynamically set based on the burial depth and geological conditions of the specific coal seam where the borehole is located, rather than using a globally fixed value, making the constraints more reasonable. Through these expansions and refinements of the model structure, loss function, and training strategy, the accuracy, robustness, and reliability of gas pressure inversion can be further improved.

[0075] Example 4: System Integration and Deployment Example The method of this invention is integrated into a workable system and deployed. This system includes the following core hardware components: Sensing layer: Explosion-proof infrared gas sensors (or thermal mass flow meters) deployed at the borehole opening for real-time, continuous acquisition of gas emission data. Q ( t ).

[0076] Data Acquisition and Input Layer: Explosion-proof data acquisition unit (or industrial PLC) is responsible for receiving sensor data and providing a human-machine interface or data interface for inputting or automatically retrieving borehole locations from a geological database. x and current coal seam parameters [ , k , a , b , s , h ].

[0077] Core computing layer: Intrinsically safe or explosion-proof edge computing terminal for mining applications, which embeds a fully trained and optimized dual-channel LSTM gas pressure inversion model. This terminal has sufficient computing power (such as based on ARM or low-power x86 architecture) to support real-time inference of the model.

[0078] Application demonstration and transmission layer: Explosion-proof local display terminal (such as touch screen) and mine gigabit industrial Ethernet communication module are used to output and transmit inversion results.

[0079] The specific workflow is as follows: After drilling and coring operations begin at the downhole drilling site, the gas sensor is immediately activated, continuously monitoring the outflow at a high frequency (e.g., 1 Hz). The data acquisition unit synchronously records the sensor data and binds and packages it with pre-entered coal seam parameters. When the monitoring time reaches the preset effective duration (e.g., 30 minutes), T When the drilling rig stops (with 1800 data points), the complete data packet is automatically sent to the edge computing terminal. The embedded model in the terminal immediately performs forward propagation calculations, typically outputting the predicted gas pressure value within seconds. The results are displayed in real-time on the explosion-proof display screen at the drilling site, in both digital and graphical formats. Audible and visual alarms are triggered based on preset thresholds, enabling on-site technicians and safety officers to make rapid decisions. Simultaneously, the data can be uploaded to the ground safety monitoring center via the industrial ring network, directly integrating into the mine's gas disaster early warning and management system platform, providing immediate and crucial data support for regional gas disaster prevention and control. The entire process achieves closed-loop automation from automatic data acquisition, intelligent processing, real-time inversion to result application, shortening the traditional gas pressure measurement work, which used to take days or even weeks, to be completed during drilling operations. This significantly improves the efficiency and timeliness of gas parameter acquisition and substantially reduces interference with normal mining operations.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for in-situ measurement of coal seam gas pressure during drilling based on LSTM, characterized in that: Includes the following steps: S1: Establish a theoretical calculation model for gas emission based on gas pressure; S2: Obtain the actual historical data information of the current coal seam in the mining area, and preprocess the data information; S3: Using the historical coal seam information data obtained in S2, perform theoretical calculations on the time-series gas emission rate, generate the corresponding time-series gas emission rate curve, and form a structured dataset. S4: Collect on-site coal seam information data and actual gas emission time-series data, perform preprocessing, and add the structured dataset obtained in S3. Gaussian noise was removed and aligned with the actual field measurement data. S5: Construct a dual-channel LSTM gas pressure inversion neural network model, and train and validate the model; S6: Embed the trained dual-channel LSTM gas pressure inversion neural network model into the explosion-proof edge computing terminal.

2. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 1, characterized in that: In S1, the theoretical calculation model for gas emission based on gas pressure is expressed as follows: in, This refers to the time-series gas emission rate. P The gas pressure of the coal seam to be measured. The porosity of the coal seam. k For coal seam permeability, a and b Let be the adsorption constant of coal. σ The effective stress coefficient of coal, h For coal seam thickness, t For time, x This is the borehole location information.

3. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 2, characterized in that: In S3, the gas emission time-series curve is calculated using the following formula: in, This refers to the amount of gas emitted from the borehole wall during the drilling process. The specific formula for calculating the gas emission rate from the borehole wall in the stopped drilling section is as follows: in, and These represent the drilling duration and the shutdown duration, respectively, in seconds (s). The instantaneous gas outburst intensity caused by coal breaking, expressed in meters (m). 3 / (m 2 ·s); and These are the geometric domains of the hole's peripheral wall and bottom wall, respectively. The Darcy velocity on the wall is expressed in m / s. k Coal permeability, in meters (m). 2 ; The viscosity is the fluid dynamic viscosity, expressed in Pa·s. The gas pressure gradient on the wall surface; The radius of the borehole is in meters. This refers to the drilling speed, expressed in m / min. The length of the borehole is in meters (m).

4. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 1, characterized in that: In S2, the actual historical data information includes geological structure data, coal sample porosity, permeability, adsorption constant, historical gas pressure and its changing trend, and effective stress coefficient of coal.

5. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 1, characterized in that: In S3, by iterating through different gas pressures Calculate and generate the corresponding time-series curve of gas emission. To form a structured dataset .

6. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 1, characterized in that: In S4, the preprocessing includes data cleaning, handling missing values ​​and outliers.

7. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 1, characterized in that: In S5, the main input variable of the dual-channel LSTM gas pressure inversion neural network model is the time series data of gas emission. ,in T The number of sampling points; the auxiliary input variable is coal seam information data, including porosity. Penetration rate k Adsorption constant a and b Effective stress coefficient and coal seam thickness h and location information x Perform normalization processing; The output variable is the coal seam gas pressure.

8. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 7, characterized in that: The dual-channel LSTM gas pressure inversion neural network model includes: The temporal feature extraction branch employs a two-layer bidirectional LSTM with 128 hidden units per layer, and the input is a gas outflow sequence. The output is a 256-dimensional hidden state vector; The prior knowledge encoding branch uses a two-layer fully connected network, with normalized coal seam parameters as input. , k , a , b , σ , h The output is a 16-dimensional encoded vector; The feature concatenation layer concatenates the hidden state vector and the encoded vector, and then outputs the predicted gas pressure value through a three-layer decoder. .

9. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 8, characterized in that: In S5, the loss function used for model training is: in, Mean squared error; Physical constraints; For boundary constraints; These are the weighting coefficients. This represents the actual stress value. Predict pressure values ​​for the model.

10. The method for in-situ measurement of coal seam gas pressure based on LSTM as described in claim 1, characterized in that: In S5, during model training, the training dataset includes the structured dataset obtained from theoretical calculations in S3 and the field test data. The structured dataset is divided into training, validation, and test sets in a 7:2:1 ratio, while the field test data is divided into training, validation, and test sets in a 4:4:2 ratio.

Images