Coal seam gas pressure and content dynamic prediction method based on multi-source data fusion

Abstract

The application discloses a coal seam gas pressure-content dynamic prediction method based on multi-source data fusion, relates to the technical field of coal seam gas prediction, and comprises the following steps: step S1, a sensor network is arranged in a downhole drilling and a mining influence area respectively, and physical field data is collected in real time and synchronously; step S2, a physical model is constructed based on Langmuir adsorption theory, and the basic content of adsorbed state gas in the coal seam is calculated; step S3, a joint correction equation fusing gas pressure and content is constructed to correct the content of adsorbed state gas in the coal seam; step S4, the corrected content of adsorbed state gas is superimposed with the content of free state gas monitored in real time, and the total content of coal seam gas is generated; and step S5, a mixed neural network model is constructed to predict a continuous curve of the coal seam gas pressure-content at a future time. The method provides a strong basis for coal seam area prediction, and further improves the efficiency of coal mining.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal seam gas prediction, and particularly relates to a coal seam gas pressure-content dynamic prediction method based on multi-source data fusion. BACKGROUND

[0002] With the continuous increase of mining depth, the influence of stress on coal and gas outburst hazard gradually increases, and low-index outburst accidents occur from time to time, becoming the No. 1 problem of deep mining in outburst coal seams. At present, with the increase of coal mining depth, coal and gas disaster accidents are one of the main disasters affecting the safety production of coal mines.

[0003] Coal and gas outburst during coal mining is generally the result of the joint action of many factors such as gas, ground stress and physical and mechanical properties of coal. Gas not only participates in the crushing of coal body, but also is the main power of throwing out the coal body, so the acquisition of the original gas content of coal seam is the basis for gas prevention and control measures. Dynamic prediction of coal seam gas pressure and content is the core technology of coal mine gas disaster prevention and control, and is directly related to the accuracy of outburst early warning and optimization of extraction efficiency.

[0004] A method and system for quickly estimating coal seam gas content based on big data are disclosed in Chinese patent CN116658244A, which includes the following steps: after drilling to the target coal seam, the drilling is withdrawn and the sealed coring device is sent to the bottom of the drill hole; the target coal seam is sampled using the sealed coring device to obtain the corresponding coal sample, and the drill hole is sealed; the coal sample is subjected to industrial analysis and mercury injection experiment to obtain the parameter information of the coal sample; after the gas pressure stabilizes, the gas pressure value of the target coal seam is read underground; the gas content of the target coal seam is estimated according to the parameter information of the coal sample and the gas pressure value of the target coal seam. The invention is used to solve the technical problems of the existing coal seam gas content prediction method, such as complicated and troublesome manual calculation, large error, easy to make mistakes, low work efficiency, etc., so as to quickly and accurately estimate the coal seam gas content.

[0005] A coal seam gas emission anomaly advanced detection and early warning method is disclosed in Chinese Patent No. CN117489413B. A correlation model of coal seam gas multi-parameters and gas emission quantity is established. During the tunneling process, a super-long directional borehole along the coal seam is constructed to measure coal seam residual gas content, coal seam residual gas pressure, coal body firmness coefficient, and drilling cuttings gas desorption index, etc. The parameter measurement results and underground roadway air volume and gas concentration sensor data are uploaded to the ground big data center in real time. The ground big data center makes intelligent analysis and prediction of gas abnormal emission in front of the working face based on the gas emission quantity model, historical database, and machine learning algorithm, and sends an alarm to the underground in a timely manner, achieving the purpose of advanced prediction and early warning of coal seam gas emission anomaly. The invention can predict and warn future gas emission anomalies in front of the working face, ensuring the rapid and safe tunneling of coal mine roadways.

[0006] The above patents all have the problems proposed in the background art: traditional methods rely on point measurement of drilling gas pressure or content, making it difficult to capture the coupling effects of coal body rupture, temperature / humidity fluctuations, etc. under mining disturbance; existing physical models (such as the Langmuir equation) are sensitive to temperature and humidity, but conventional correction methods only use fixed coefficients and do not establish a dynamic correlation mechanism with coal quality characteristics. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a coal seam gas pressure-content dynamic prediction method based on multi-source data fusion to overcome the shortcomings of the prior art.

[0008] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0009] The coal seam gas pressure-content dynamic prediction method based on multi-source data fusion comprises the following steps:

[0010] Step S1, deploy sensor networks in underground boreholes and mining affected areas to collect physical field data in real time and synchronously;

[0011] Step S2, construct a physical model based on the Langmuir adsorption theory to calculate the basic content of adsorbed coal seam gas;

[0012] Step S3, construct a joint correction equation that fuses gas pressure and content to correct the adsorbed coal seam gas content;

[0013] Step S4, superimpose the corrected adsorbed gas content and the real-time monitored free gas content to generate the total coal seam gas content;

[0014] Step S5, construct a hybrid neural network model to predict and generate the continuous curve of coal seam gas pressure-content at future time.

[0015] Further, the step S1 specifically comprises: setting a distributed optical fiber pressure sensor inside the borehole to collect the gas pressure distribution along the axial direction of the borehole, setting a spectral gas concentration sensor to obtain the volume concentration of free state gas in real time, and setting an acoustic emission sensor array to capture the acoustic signals generated by the micro-fracture of the coal body; setting a ground stress sensor in the mining area to collect and measure the vertical ground stress changes, setting a temperature and humidity sensor to monitor the environmental temperature and humidity parameters, and setting a high-precision barometer to record the atmospheric pressure fluctuations in the mining area.

[0016] Further, in the step S2, the specific formula of the coal seam adsorbed gas basic content is:

[0017]

[0018] wherein, represents the coal seam adsorbed gas content, and respectively represent the temperature correction factor and the humidity correction factor, represents the coal density, and respectively represent the Langmuir volume constant and the Langmuir pressure constant, represents the real-time collected gas pressure.

[0019] Further, the calculation formula of the temperature correction factor is:

[0020]

[0021] wherein, represents the temperature coefficient, represents the environmental temperature, represents the reference temperature.

[0022] wherein, the calculation formula of the temperature coefficient is:

[0023]

[0024] wherein, represents the adsorption activation energy.

[0025] Further, the calculation formula of the humidity correction factor is:

[0026]

[0027] wherein, represents the environmental humidity, represents the humidity coefficient.

[0028] wherein, the calculation formula of the humidity coefficient is:

[0029]

[0030] wherein, represents the fixed carbon content in coal.

[0031] Further, in the step S3, the specific formula of the combined correction equation is:

[0032]

[0033] wherein, represents the corrected coal seam gas content, represents the pressure change feedback coefficient.

[0034] Further, in the combined correction equation, the calculation formula of the pressure change feedback coefficient is:

[0035]

[0036] wherein, represents the symbol function, represents the cumulative energy of acoustic emission.

[0037] Further, in the step S4, the calculation formula of the total coal seam gas content is:

[0038]

[0039] wherein, represents the collected free gas concentration.

[0040] Further, the step S5 specifically includes the following steps:

[0041] Step S5.1, inputting the historical time series data and real-time state vector into the LSTM neural network to obtain a hidden state sequence;

[0042] Step S5.2, inputting the hidden state sequence into the TCN network as a time series;

[0043] Step S5.3, the TCN network outputs to generate a predicted coal seam gas pressure and content sequence;

[0044] Step S5.4, fitting the predicted coal seam gas pressure and content sequence to obtain a continuous curve of coal seam gas pressure-content;

[0045] wherein, the TCN network includes three TCN layers, each layer structure is empty causal convolution, ReLU activation, layer normalization and residual connection, the convolution kernel size is 9, the channel number is 128, 64 and 32 respectively, and the hole rate is 1, 2 and 4 respectively;

[0046] The historical time series data includes corrected coal seam gas pressure and content, ground stress and acoustic emission energy at a past 60-second time point;

[0047] The real-time state vector includes real-time temperature, humidity, air pressure, coal seam gas pressure spatial gradient and coal seam gas pressure change rate absolute value at a current time point.

[0048] Further, in the step S5.4, the predicted coal seam gas pressure sequence and the predicted coal seam gas content sequence output by the TCN network are respectively subjected to cubic spline interpolation fitting processing, to generate smooth and continuous coal seam gas pressure-time curve and coal seam gas content-time curve in a future time period, which jointly constitute the coal seam gas pressure-content continuous curve.

[0049] Compared with the prior art, the present application has the following beneficial effects:

[0050] 1. The present application improves the prediction accuracy by multi-source fusion, and sets distributed optical fiber pressure sensors, spectral gas concentration sensors, acoustic emission sensors in the borehole, and sets ground stress / temperature / humidity / air pressure sensors in the mining area, to construct a full-dimensional monitoring network.

[0051] 2. The present application generates a temperature / humidity correction factor based on the adsorption activation energy and the fixed carbon content of coal quality characteristics to replace the traditional fixed coefficient, and drives a pressure change feedback coefficient through acoustic emission cumulative energy, to real-time correct the adsorption model, so that the calculated adsorbed gas content is more accurate.

[0052] 3. The present application adopts LSTM-TCN dual network architecture: LSTM extracts long time series correlation (ground stress / air pressure gradient), and TCN captures local mutation (acoustic emission / pressure change rate), which can more effectively capture the characteristics of multi-source data and more accurately obtain the prediction result.

[0053] 4. The present application realizes gas dynamic coupling analysis and minute-level continuous early warning capability through the "multi-source sensing-physical correction-damage feedback-intelligent prediction" closed-loop architecture, breaks through the bottleneck of the prior art, and provides high reliability guarantee for intelligent mine gas disaster prevention and control. BRIEF DESCRIPTION OF DRAWINGS

[0054] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0055] Figure 1 is a flowchart of an embodiment of the present application;

[0056] Figure 2 is a prediction flowchart of an embodiment of the present application;

[0057] Figure 3 A TCN network architecture diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be described in detail below with reference to the drawings and specific embodiments.

[0059] As Figure 1 shown, the coal seam gas pressure-content dynamic prediction method based on multi-source data fusion includes the following steps:

[0060] Step S1, deploying sensor networks in underground boreholes and mining-affected areas respectively, and collecting physical field data in real time and synchronously;

[0061] Step S2, constructing a physical model based on Langmuir adsorption theory, and calculating the basic content of adsorbed gas in the coal seam;

[0062] Step S3, constructing a joint correction equation fusing gas pressure and content to correct the adsorbed gas content in the coal seam;

[0063] Step S4, superimposing the corrected adsorbed gas content and the real-time monitored free gas content to generate the total coal seam gas content;

[0064] Step S5, constructing a hybrid neural network model to predict and generate the continuous curve of coal seam gas pressure-content at future time.

[0065] The step S1 specifically includes: setting a distributed optical fiber pressure sensor inside the borehole to collect the axial gas pressure distribution along the borehole, setting a spectral gas concentration sensor to obtain the real-time free gas volume concentration, and setting an acoustic emission sensor array to capture the acoustic signals generated by the micro-fracture of the coal body; setting a ground stress sensor in the mining area to collect and measure the vertical ground stress changes, setting a temperature and humidity sensor to monitor the environmental temperature and humidity parameters, and setting a high-precision barometer to record the atmospheric pressure fluctuations in the mining area.

[0066] The acoustic emission sensor is made by using the principle that the physical properties of certain substances (such as semiconductors, ceramics, piezoelectric crystals, strong magnetic bodies and superconductors, etc.) change with the external measured quantity. It utilizes many effects (including physical, chemical and biological effects) and physical phenomena, such as using the piezoresistance, humidity, heat, light, magnetic and gas sensitivity effects of materials, to convert the measured quantities such as strain, humidity, temperature, displacement, magnetic field and gas into electric quantities.

[0067] In the step S2, the specific formula of the basic content of adsorbed gas in the coal seam is:

[0068]

[0069] wherein, represents the adsorbed gas content of the coal seam, and respectively represent a temperature correction factor and a humidity correction factor, represents the density of the coal body, and respectively represent a Langmuir volume constant and a Langmuir pressure constant, represents the real-time collected gas pressure.

[0070] wherein, the source of each parameter data is shown in Table 1:

[0071] Table 1

[0072]

[0073] wherein, the density of the coal body is measured based on a laboratory method, determined by a true density instrument:

[0074] Typical values include: anthracite: 1.4~1.8 t / m³, lignite: 1.1~1.3 t / m³.

[0075] The calculation formula of the temperature correction factor is:

[0076]

[0077] wherein, represents a temperature coefficient, represents the ambient temperature, represents a reference temperature;

[0078] wherein, the calculation formula of the temperature coefficient is:

[0079]

[0080] wherein, represents an adsorption activation energy, determined by a TGA-DSC combined instrument.

[0081] Physical nature of the adsorption activation energy: the adsorption activation energy is the energy barrier that needs to be overcome for the gas molecules to change from the free state to the adsorbed state, reflecting the adsorption strength of the coal to the gas; it can be fitted by measuring the change of adsorption capacity at different temperatures by a thermogravimetric-differential scanning calorimeter (TGA-DSC).

[0082] The calculation formula of the humidity correction factor is:

[0083]

[0084] wherein, represents the ambient humidity, represents a humidity coefficient;

[0085] The calculation formula of the humidity coefficient is:

[0086]

[0087] wherein, represents the fixed carbon content in coal, which is obtained by industrial analysis.

[0088] In the step S3, the specific formula of the joint correction equation is:

[0089]

[0090] wherein, represents the corrected coal seam gas content, represents the pressure change feedback coefficient, represents the real-time collected gas pressure.

[0091] In the joint correction equation, the calculation formula of the pressure change feedback coefficient is:

[0092]

[0093] wherein, represents the sign function, represents the cumulative energy of acoustic emission.

[0094] The calculation formula of the total coal seam gas content is:

[0095]

[0096] wherein, represents the collected free gas concentration.

[0097] As shown in Figure 2 the step S5 specifically comprises the following steps:

[0098] Step S5.1, input the historical time series data and real-time state vector into the LSTM neural network to obtain a hidden state sequence;

[0099] Step S5.2, input the hidden state sequence into the TCN network as a time series;

[0100] Step S5.3, the TCN network outputs to generate a predicted coal seam gas pressure and content sequence;

[0101] Step S5.4, fitting the predicted coal seam gas pressure and content sequence to obtain a continuous curve of coal seam gas pressure-content;

[0102] Wherein, wherein, the TCN network comprises three TCN layers, each layer structure is a hollow causal convolution, ReLU activation, layer normalization and residual connection, the convolution kernel size is 9, the channel number is 128, 64 and 32 respectively, and the hollow rate is 1, 2 and 4 respectively;

[0103] The historical time series data includes the corrected coal seam gas pressure and content, ground stress and acoustic emission energy at the past 60-second time points;

[0104] The real-time state vector includes the real-time temperature, humidity, air pressure, coal seam gas pressure spatial gradient and coal seam gas pressure change rate absolute value at the current time point.

[0105] In the step S5.4, the predicted coal seam gas pressure sequence and the predicted coal seam gas content sequence output by the TCN network are respectively fitted by using the cubic spline interpolation method, to generate smooth and continuous coal seam gas pressure-time curve and coal seam gas content-time curve in the future time period, which jointly constitute the coal seam gas pressure-content continuous curve.

[0106] The dual-channel input structure includes a historical data channel and a real-time state channel:

[0107] The historical data channel receives 4-dimensional time series data (pressure, concentration, stress, acoustic emission energy) in a 60-second time window, with a sampling rate of 1 Hz, to form a 60x4 matrix.

[0108] The real-time state channel receives a 5-dimensional physical feature vector (temperature, humidity, air pressure, pressure spatial gradient, pressure change rate)

[0109] The LSTM encoder adopts a double-layer LSTM structure, with 128 hidden units in each layer, processes the historical data tensor, and calculates along the time step to finally output the hidden state vector (dimension 128) of each time step, to finally obtain 60 time step hidden state sequences.

[0110] Bidirectional hierarchical processing: 2-layer bidirectional LSTM structure, 128 units per layer:

[0111] The first layer extracts short-term fluctuation features (<5 seconds period), and the second layer captures long-term trend features (>30 seconds period), the interlayer residual connection avoids gradient disappearance, and the Dropout layer (rate=0.2) prevents overfitting.

[0112] The main structure of the TCN network includes the convolution kernel and the final output pressure prediction:

[0113] Convolution kernel: width 9, expansion factor [1, 2, 4];

[0114] Final output pressure prediction: 30-dimensional vector (corresponding to the next 30 seconds) and content prediction: 30-dimensional vector (corresponding to the next 30 seconds).

[0115] Wherein the specific dimensions and parameters of the LSTM neural network and the TCN network are shown in Table 2:

[0116] Table 2

[0117]

[0118] LSTM is a special type of recurrent neural network designed to address the problem of long-term sequence data dependencies. Traditional RNNs suffer from gradient vanishing or exploding problems when dealing with long sequence data, making it difficult to effectively learn long-term dependencies. LSTM addresses this issue by introducing a gating mechanism.

[0119] LSTM consists of four key components: input gate, forget gate, output gate, and cell state. Each component has a different role and works together to transfer and control information.

[0120] Input gate: controls the amount of new information input into the cell state. The input gate consists of a sigmoid activation function and a dot product operation, which determines which information will be updated. Forget gate: determines which information should be forgotten or deleted. The forget gate consists of a sigmoid activation function and a dot product operation, which controls which information in the updated cell state is ignored. Output gate: controls the flow of information from the cell state to the output. The output gate consists of a sigmoid activation function and a dot product operation, which filters the information in the cell state and outputs it to the next layer. Cell state: the core component that runs throughout the entire LSTM, used to capture and transfer long-term dependency information. The cell state can be updated and adjusted through the operations of the input gate and the forget gate.

[0121] LSTM can effectively handle long-term dependency problems, avoiding the problems of gradient vanishing or exploding, and improving the model's performance on time series data and sequence data.

[0122] As shown in Figure 3 TCN is a network structure specifically designed to handle time series data by using convolutional layers instead of recurrent layers to handle sequence dependencies. The key features of TCN include:

[0123] Causal convolution: ensures that only data up to the current time is used when predicting the value at the current time, ensuring the causality of the model. Dilated convolution: expands the receptive field of the convolutional layer, allowing the network to capture long-range sequence dependencies without increasing the number of parameters or computational complexity.

[0124] The specific parameters of each layer in the three layers of the TCN network are shown in table 3.

[0125] Table 3

[0126]

[0127] The cubic spline interpolation is a method for constructing a smooth curve by piecewise cubic polynomial, which can accurately pass through the given data points and keep the second derivative of the curve continuous. The core is to construct the interpolation function by using three bending moments or three slope methods, and solve the equation set combined with the boundary conditions, which is widely used in mathematics, engineering and other fields. By predicting the coal seam gas pressure and content and constructing a smooth curve, the predicted data can be visualized, so that the coal seam early warning is more intuitive.

[0128] The examples described in the application are only used to describe the preferred embodiments of the application, and do not limit the concept and scope of the application. Without departing from the design idea of the application, various deformations and improvements of the technical solutions of the application made by the engineering and technical personnel in the field shall fall within the protection scope of the application.

Claims

1. A coal seam gas pressure-content dynamic prediction method based on multi-source data fusion, characterized in that, Comprise the following steps: Step S1, respectively deploying sensor networks in the downhole borehole and mining influence area, real-time synchronous acquisition of physical field data; Step S2, based on Langmuir adsorption theory, a physical model is constructed to calculate the basic content of adsorbed gas in coal seam; Step S3, a joint correction equation combining gas pressure and content is constructed to correct the adsorbed gas content in coal seam; Step S4, superimpose the corrected adsorbed gas content and the real-time monitored free gas content to generate the total coal seam gas content; Step S5, a hybrid neural network model is constructed to predict the continuous curve of coal seam gas pressure-content at future time; In the step S2, the specific formula of the basic content of adsorbed gas in coal seam is: ； wherein, represents the adsorbed gas content of the coal seam, and respectively represent a temperature correction factor and a humidity correction factor, represents the density of the coal body, and respectively represent the Langmuir volume constant and the Langmuir pressure constant; The calculation formula of the temperature correction factor is: ； wherein, represents a temperature coefficient, represents an ambient temperature, represents a reference temperature; The calculation formula of the temperature coefficient is: ； wherein, represents the adsorption activation energy; The calculation formula of the humidity correction factor is: ； wherein represents the ambient humidity, represents the humidity coefficient; The calculation formula of the humidity coefficient is: ； wherein, represents the fixed carbon content in the coal; In the step S3, the specific formula of the joint correction equation is: ； wherein, represents the corrected coal seam gas content, represents the pressure change feedback coefficient, represents the real-time collected gas pressure; In the joint correction equation, the calculation formula of the pressure change feedback coefficient is: ； wherein denotes the Heaviside function, denotes the cumulative energy of the acoustic emission.

2. The method of claim 1, wherein, The step S1 specifically includes: setting a distributed optical fiber pressure sensor inside the borehole to collect the axial gas pressure distribution along the borehole, setting a spectral gas concentration sensor to obtain the real-time free gas volume concentration, and setting an acoustic emission sensor array to capture the acoustic signals generated by the micro-fracture of the coal body; In the mining area, a ground stress sensor is arranged to collect and measure the vertical ground stress change, a temperature and humidity sensor is arranged to monitor the environmental temperature and humidity parameters, and a high-precision barometer is arranged to record the atmospheric pressure fluctuation in the mining area.

3. The method of claim 2, wherein, In the step S4, the calculation formula of the total coal seam gas content is: ； wherein, represents the concentration of free gas collected.

4. The method of claim 3, wherein, The step S5 specifically includes the following steps: Step S5.1, input the historical time series data and real-time state vector into the LSTM neural network to obtain the hidden state sequence; Step S5.2, input the hidden state sequence into the TCN network as a time series; Step S5.3, the TCN network outputs to generate the predicted coal seam gas pressure and content sequence; Step S5.4, fitting the predicted coal seam gas pressure and content sequence to obtain the continuous curve of coal seam gas pressure-content; The TCN network includes three TCN layers, each layer structure is empty causal convolution, ReLU activation, layer normalization and residual connection, the convolution kernel size is 9, the channel number is 128, 64 and 32 respectively, and the hole rate is 1, 2 and 4 respectively; The historical time series data includes: the corrected coal seam gas pressure and content, ground stress and acoustic emission energy at the past 60 second time points; The real-time state vector includes: real-time temperature, humidity, air pressure, coal seam gas pressure spatial gradient and coal seam gas pressure change rate absolute value at the current time point.

5. The method of claim 4, wherein, In the step S5.4, the predicted coal seam gas pressure sequence and the predicted coal seam gas content sequence output by the TCN network are respectively fitted by the cubic spline interpolation method to generate smooth and continuous coal seam gas pressure-time curve and coal seam gas content-time curve in the future time period, which together constitute the continuous curve of coal seam gas pressure-content.

Images