Gas extraction and roadway environment parameter collaborative prediction system and method

Abstract

The invention discloses a gas extraction and roadway environment parameter collaborative prediction system and method, belongs to the technical field of coal mine gas extraction and safety monitoring, and aims at improving the precision and accuracy of gas concentration prediction and improving the working face gas extraction efficiency and safety. The system comprises a wind speed acquisition system, a fire beam tube monitoring system, a high-position drilling gas extraction monitoring system, a corner gas concentration monitoring system, a PLC (programmable logic controller) and a computer. A partial differential physical equation and related information of boundary conditions and initial value conditions of the partial differential physical equation are integrated into a loss function of a full-connection neural network, a neural network model approaching a partial differential equation solution is obtained through learning, a mapping relation between gas extraction parameters and extraction effects is established, flow characteristics of gas in a coal mine tunnel are accurately simulated and predicted, and the gas extraction efficiency is improved. And gas over-limit early warning can be carried out, so that a guarantee is provided for safe production of a stope face.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal mine gas drainage and safety monitoring, and specifically relates to a system and method for collaborative prediction of gas drainage and roadway environment parameters. Background Technique

[0002] The prevention and control of mine gas is the key to ensuring the safe and efficient mining of mines. With the increase in the depth of coal mining and the improvement of mining intensity and efficiency, the problems of gas overrun in the upper corner and return air current of the working face have become increasingly serious. The "hole instead of roadway" gas drainage technology in the gob area by high-level boreholes can achieve stable and efficient control of gas drainage and effectively solve the problem of gas overrun in the upper corner. Therefore, mastering the real-time change law of the gas drainage effect in the gob area of high-level boreholes, including the cumulative gas production or concentration of the drained gas and the gas concentration in the upper corner, is particularly important for the selection of gas drainage parameters and the evaluation of production safety.

[0003] At present, the existing methods for predicting coal mine gas concentration mainly analyze the characteristics and laws of gas concentration time series through nonlinear theory, but the defects mainly include:

[0004] 1. Lack of mechanism-data fusion: The existing methods overly rely on data statistical characteristics and ignore the gas migration dynamics mechanism and the evolution law of the mining-induced fracture field, resulting in insufficient interdisciplinary integration of mechanism modeling and data-driven approaches;

[0005] 2. Absence of multi-parameter coupling prediction: For the gas drainage scenario of high-level boreholes, an associated prediction model for multi-source parameters such as the gas concentration field in the upper / lower corner and the ratio of drainage flow to concentration has not been constructed. Summary of the Invention

[0006] Aiming at the above-mentioned existing technical deficiencies, the purpose of the present invention is to provide a collaborative prediction system and method for gas drainage and roadway environment parameters that integrates mechanisms, improve the accuracy and precision of gas concentration prediction, and increase the gas drainage efficiency and safety of the working face.

[0007] To solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A system for collaborative prediction of gas drainage and roadway environment parameters includes a wind speed acquisition system, a fire bundle tube monitoring system, a high-level borehole gas drainage monitoring system, a corner gas concentration monitoring system, a PLC programmable logic controller, and a computer, wherein:

[0009] The wind speed acquisition system is used to measure the average wind speed of the full cross-section of the roadway at the positions of detection points A to D in real time. The wind speed acquisition system includes a transportation gate road, a return air gate road, an upper corner, and a lower corner. A mining passage is arranged between the coal mining face and the goaf. The head and tail ends of the mining passage are respectively connected to the transportation gate road and the return air gate road. The connection between the transportation gate road and the mining passage is set as the lower corner. The A detection point is set on the side of the lower corner close to the transportation gate road, and the B detection point is set on the side of the lower corner close to the mining passage; the connection between the return air gate road and the mining passage is set as the upper corner. The D detection point is set on the side of the upper corner close to the return air gate road, and the C detection point is set on the side of the upper corner close to the mining passage. Ultrasonic wind speed measuring devices are respectively arranged at the positions of the A detection point, the B detection point, the C detection point, and the D detection point;

[0010] The fire beam tube monitoring system is used to measure the gas concentration at different positions in the goaf in real time. The fire beam tube monitoring system includes a number of upper corner fire beam tubes, a number of lower corner fire beam tubes, and a fire beam tube analysis device. Along the length direction of the cross-section of the goaf on the side close to the upper corner, a1, a2, a3... a n upper corner fire beam tubes are arranged in sequence. Along the length direction of the cross-section of the goaf on the side close to the lower corner, b1, b2, b3... b n lower corner fire beam tubes are arranged in sequence. The fire beam tube analysis device is electrically connected to a number of upper corner fire beam tubes and lower corner fire beam tubes respectively;

[0011] The high-level borehole gas drainage monitoring system is used to monitor the gas drainage flow rate and CH4 concentration of the high-level borehole in real time. The high-level borehole gas drainage monitoring system includes a drainage pump and a number of high-level borehole pipelines. The drainage pump is located on the side of the transportation gate road. A mass flow meter, a No. 1 concentration sensor, and a pressure sensor are respectively arranged on the drainage pump; One end of the high-level borehole pipeline extends into the goaf, and the other end of the high-level borehole pipeline is connected to the drainage pump;

[0012] The corner gas concentration monitoring system is used to monitor the gas concentration at the positions of the upper corner and the lower corner in real time. The corner gas concentration monitoring system includes a No. 2 concentration sensor arranged at the position of the upper corner and a No. 3 concentration sensor arranged at the position of the lower corner;

[0013] A PLC programmable logic controller is set in the computer. The data signals collected by the wind speed acquisition system, the fire beam tube monitoring system, the high-level borehole gas drainage monitoring system, and the corner gas concentration monitoring system are respectively connected to the signal input end of the PLC programmable logic controller through signal lines. The PLC programmable logic controller saves the data content to the computer and processes the data output by the PLC through the software installed on the computer, and is then used for the collaborative prediction of gas drainage and roadway environment parameters.

[0014] A method for collaborative prediction of gas drainage and roadway environment parameters using the system described above, comprising the following steps:

[0015] S1. Use the collaborative prediction system for gas drainage and roadway environment parameters to collect gas and fluid mechanics parameters at different drainage times in the mining face, and build a scatter observation data set A1(A 11 ,A 12 ,A 13 );

[0016] Wherein:

[0017] The data set A 11 (X 11 ,t 11 ,p 11 ,u 11 ,c 11 ) includes the position coordinates X 11 of the drainage hole and the CH4 drainage time t 11 ,drainage pressure p 11 ,drainage flow u 11 ,drainage concentration c 11 ;

[0018] The data set A 12 (X 12 ,t 12 ,p 12 ,u 12 ,c 12 ) includes the position coordinates X 12 of different measurement points at the upper corner and the lower corner and the drainage time t 12 ,drainage pressure p 12 ,roadway wind speed u 12 ,CH4 concentration c 12 ;

[0019] The data set A 13 (X 13 ,t 13 ,c 13 ) is the CH4 concentration c 13 measured by the fire bundle tube monitoring system at different positions X 13 in the goaf at different times t 13 ;

[0020] When the geometric model is in a two-dimensional state, the coordinates of X 11 ,X 12 ,X 13 are (x, y); when the geometric model is in a three-dimensional state, the coordinates of X 11 ,X 12 ,X 13 are (x, y, z);

[0021] S2. Use computational fluid dynamics software to simulate and calculate the flow characteristics of gas in coal mine tunnels, solve the problem by setting initial conditions and boundary conditions, and obtain the X(x i 、x j )’s velocity u, pressure p and concentration c, and construct the numerical calculation result data set A2(X, t, p, u, c);

[0022] in:

[0023] The gas flow in the goaf satisfies the turbulence equation, continuity equation and momentum conservation equation. The expression F1 is:

[0024]

[0025] In formula (1), ρ is the gas density, unit is kg / m 3 ; is a universal variable; t is time; Γ is the generalized diffusion coefficient; S is the generalized source term; x i 、x j are the coordinates of the three basic directions in three-dimensional coordinates, i = 1, 2 or 3, j = 1, 2 or 3;

[0026] The gas diffusion equation F2 is:

[0027]

[0028] In formula (2), c0 is the gas concentration; u is the gas velocity; S i is the discrete phase conversion rate; D is the diffusion coefficient of gas; ▽ represents the Hamiltonian operator;

[0029] Boundary conditions:

[0030]

[0031] In formula (3), Φ B is the boundary condition function; Γ i represents the i-th boundary; p i , c i ,u i represents the pressure, concentration, and velocity of the i-th boundary condition; f(p i , c i ,u i ) represents the function value of the i-th boundary condition;

[0032] Initial conditions:

[0033]

[0034] In formula (4), Φ I is the initial condition function;i represents the i-th initial value condition region; p i , c i , u i respectively represent the pressure, concentration, and velocity of the i-th initial value condition; g(p i , c i , u i ) represents the function value of the i-th initial value condition;

[0035] S3. Data preprocessing: Combine the scatter observation data set A1 (A 11 , A 12 , A 13 ) constructed in step S1 and the numerical calculation result data set A2 (X, t, p, u, c) constructed in step S2, and perform preprocessing, including data cleaning and normalization operations, to form the data set A0, and divide it into a training set and a test set;

[0036] S4. Deep learning model training: Use the multi-layer fully connected neural network algorithm (MLP) to train the training set of the gas drainage system, and the outputs are p, u, and c. The activation function uses the smooth tanh function to ensure the existence and continuity of the derivative; then, use the TensorFlow automatic differentiation technology and adopt the tape.gradient function to find the first-order and second-order derivatives of p, u, and c with respect to time and space coordinates, avoiding the complexity of manual derivation;

[0037] Construct the total loss function:

[0038] L 总 = λ1L1 + λ PDE L PDE + λ BC L BC + λ IC L IC ; (5)

[0039] In formula (5), L1 is the loss function for data set training and is the MSE error; L PDE represents the partial differential equation residual loss; L BC represents the boundary condition loss; L IC represents the initial value condition loss; λ1, λ PDE , λ BC , λ IC are hyperparameters.

[0040] ① The partial differential equation residual loss L PDE is the mean square error of the partial differential equation residual, as follows:

[0041] L PDE = mean((F1 + F2) 2 ); (6)

[0042] ② Boundary condition loss L BC :

[0043] L BC = mean(Φ B 2 ); (7)

[0044] ③ Initial value condition loss L BC :

[0045] L BC = mean(ΦI 2 ); (8)

[0046] In equations (6) to (8), mean is the mean function;

[0047] S5. Adjust the weights and biases of the model through the gradient descent optimization algorithm, and solve Arg min L 总 , until the prediction accuracy of the model converges, and obtain the neural network link weight parameters and the physical parameters of the partial differential equation;

[0048] S6. Use the trained model to evaluate the test set constructed in step S5, and the evaluation indicators include accuracy, precision, etc.; if the evaluation result is unqualified, repeat steps S3 to S5 until the test set passes the evaluation, which is the optimized model;

[0049] S7. Use the optimized model to jointly predict the gas concentration and gas drainage parameters at the upper corner position;

[0050] S8. Gas overlimit warning: Adopt a single-index warning method, set the warning gas concentration to 1.0%, and when the predicted value of the gas concentration at the upper corner position reaches 95% of the limit value, give an early warning.

[0051] The beneficial effects of the present invention are as follows: The present invention combines the collected scattered field data with the data set solved according to the physical equation. First, the relevant information of the partial differential physical equation and its boundary conditions and initial value conditions is incorporated into the loss function of the fully connected neural network, and then the fully connected neural network is used to approximate the solution of the equation, so as to learn and obtain a neural network model that approximates the solution of the partial differential equation, establish a mapping relationship between the gas drainage parameters and the drainage effect, and can accurately simulate and predict the flow characteristics of gas in the coal mine roadway, improving the efficiency and safety of gas drainage. This method can be widely applied to the field of coal mine gas drainage, and is of great significance for improving the coal mine production efficiency and ensuring the safety of miners' lives. Description of the Drawings

[0052] Figure 1 Schematic diagram of the collaborative prediction system structure of gas drainage and roadway environment parameters for the fusion mechanism

[0053] Figure 1 In it, 1 is the coal mining face, 2 is the upper corner, 3 is the gob area, 4 is the computer, 5 is the high-level borehole pipeline, 6 is the lower corner, 7 is the transportation gateway, 8 is the return airway, 9 is the gas extraction pump, 10 is the mass flowmeter, 11 is the 1# concentration sensor, 12 is the pressure sensor, 13 is the 2# concentration sensor, 14 is the 3# concentration sensor, 15 is the upper corner fire sampling tube, 16 is the lower corner fire sampling tube, 17 is the fire sampling tube analysis device, and 18 is the PLC programmable logic controller;

[0054] Figure 2 It is the flow chart of the collaborative prediction method for gas extraction and roadway environment parameters of the fusion mechanism. Specific implementation mode

[0055] The present invention will be further described in detail below with reference to the drawings and embodiments.

[0056] As Figure 1 shown, a system for collaborative prediction of gas extraction and roadway environment parameters includes a wind speed acquisition system, a fire sampling tube monitoring system, a high-level borehole gas extraction monitoring system, a corner gas concentration monitoring system, a PLC programmable logic controller 18 and a computer 4, wherein:

[0057] The wind speed acquisition system is used to measure the average wind speed of the whole roadway section at the positions of detection points A to D in real time. The wind speed acquisition system includes the transportation gateway 7, the return airway 8, the upper corner 2 and the lower corner 6. A mining passage is arranged between the coal mining face 1 and the gob area 3. The head and tail ends of the mining passage are respectively connected to the transportation gateway 7 and the return airway 8. The connection between the transportation gateway 7 and the mining passage is set as the lower corner 6. The side of the lower corner 6 close to the transportation gateway 7 is set as detection point A, and the side of the lower corner 6 close to the mining passage is set as detection point B; the connection between the return airway 8 and the mining passage is set as the upper corner 2. The side of the upper corner 2 close to the return airway 8 is set as detection point D, and the side of the upper corner 2 close to the mining passage is set as detection point C. Ultrasonic wind speed measuring devices are arranged at the positions of detection points A, B, C and D respectively;

[0058] The fire sampling tube monitoring system is used to measure the gas concentration at different positions in the gob area in real time. The fire sampling tube monitoring system includes a number of upper corner fire sampling tubes 15, a number of lower corner fire sampling tubes 16 and a fire sampling tube analysis device 17. Along the length direction of the cross section of the gob area 3 on the side close to the upper corner 2, a1, a2, a3... a n upper corner fire sampling tubes 15 are arranged in sequence. Along the length direction of the cross section of the gob area 3 on the side close to the lower corner 6, b1, b2, b3... b nThe lower corner fire sampling tube 16 and the fire sampling tube analysis device 17 are respectively electrically connected to a plurality of upper corner fire sampling tubes 15 and the lower corner fire sampling tube 16;

[0059] The high-level borehole gas drainage monitoring system is used to monitor the flow rate and CH4 concentration of high-level borehole gas drainage in real time. The high-level borehole gas drainage monitoring system includes a drainage pump 9 and a plurality of high-level borehole pipelines 5. The drainage pump 9 is located on one side of the transportation gateway 7. A mass flow meter 10, a 1# concentration sensor 11, and a pressure sensor 12 are respectively arranged on the drainage pump 9; One end of the high-level borehole pipeline 5 extends into the goaf 3, and the other end of the high-level borehole pipeline 5 is communicated with the drainage pump 9;

[0060] The corner gas concentration monitoring system is used to monitor the gas concentration at the positions of the upper corner 2 and the lower corner 6 in real time. The corner gas concentration monitoring system includes a 2# concentration sensor 13 arranged at the position of the upper corner 2 and a 3# concentration sensor 14 arranged at the position of the lower corner 6;

[0061] A PLC programmable logic controller 18 is arranged in the computer 4. The data signals collected by the wind speed acquisition system, the fire sampling tube monitoring system, the high-level borehole gas drainage monitoring system, and the corner gas concentration monitoring system are respectively connected to the signal input end of the PLC programmable logic controller 18 through signal lines. The PLC programmable logic controller 18 saves the data content to the computer 4 and processes the data output by the PLC through the software installed on the computer 4, and is then used for the collaborative prediction of gas drainage and roadway environment parameters.

[0062] In this embodiment, the two-dimensional goaf gas drainage is used as the application scenario. As Figure 2 shown, a method for collaborative prediction of gas drainage and roadway environment parameters using the above-mentioned system includes the following steps:

[0063] S1. Use the collaborative prediction system of gas drainage and roadway environment parameters to collect the gas and fluid mechanics parameters at different drainage times of the coal mining face 1, and build a scatter observation data set A1 (A 11 , A 12 , A 13 );

[0064] Among them:

[0065] The data set A 11 (X 11 , t 11 , p 11 , u 11 , c 11 ) includes the position coordinates X 11 of the drainage hole and the CH4 drainage time t 11 of the drainage pump 9, the drainage pressure p 11, gas drainage flow rate u 11 , gas drainage concentration c 11 ;

[0066] Dataset A 12 (X 12 , t 12 , p 12 , u 12 , c 12 ) includes the position coordinates X of different measurement points at the positions of the upper corner 2 and the lower corner 6 12 and the gas drainage time t 12 , gas drainage pressure p 12 , roadway air velocity u 12 , CH4 concentration c 12 ;

[0067] Dataset A 13 (X 13 , t 13 , c 13 ) is the CH4 concentration c measured at different positions X of the goaf 3 at different times t 13 by the fire bundle tube monitoring system 13 ; 13 ;

[0068] When the geometric model is in two-dimensional state, the coordinates of X 11 , X 12 , X 13 are (x, y); when the geometric model is in three-dimensional state, the coordinates of X 11 , X 12 , X 13 are (x, y, z);

[0069] S2. Use computational fluid dynamics software to simulate and calculate the flow characteristics of gas in a coal mine roadway, control the expressions F1 of the turbulent flow equation, continuity equation and momentum conservation equation, and the gas diffusion equation F2, and solve according to the actual working conditions on site, the set initial conditions and boundary conditions, and obtain the velocity u, pressure p and concentration c at each position X(x i , x j ) at time t, and construct a numerical calculation result dataset A2(X, t, p, u, c);

[0070] S3. Data preprocessing: Combine the scatter observation dataset A1(A 11 , A 12 , A 13 ) constructed in step S1 and the numerical calculation result dataset A2(X, t, p, u, c) constructed in step S2, and perform preprocessing, including data cleaning and normalization operations, to form dataset A0;

[0071] S4. Deep learning model training: Using the multi-layer fully connected neural network algorithm (MLP), train the training set of the gas drainage system, and the outputs are p, u, and c. The activation function uses a smooth tanh function to ensure the existence and continuity of the derivative. Using the TensorFlow automatic differentiation technique, use the tape.gradient function to find the first and second derivatives of p, u, and c with respect to time and spatial coordinates, and then solve the total loss function L according to equations (5) to (8). 总 ;

[0072] S5. Adjust the weights and biases of the model through the gradient descent optimization algorithm to solve Arg minL 总 , until the prediction accuracy of the model reaches convergence, and obtain the neural network link weight parameters and the physical parameters of the partial differential equation;

[0073] S6. Use the trained model to evaluate the test set constructed in step S5. The evaluation indicators include accuracy, precision, etc. If the evaluation result is unqualified, repeat steps S3 to S5 until the test set passes the evaluation, which is the optimized model;

[0074] S7. Use the optimized model to jointly predict the gas concentration and gas drainage parameters at the upper corner 2 position;

[0075] S8. Gas overrun warning: Adopt a single-index warning method, set the warning gas concentration to 1.0%, and when the predicted value of the gas concentration at the upper corner 2 position reaches 95% of the limit value, give an early warning.

[0076] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the technical field within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claimed rights.

Claims

1. A system for collaborative prediction of gas drainage and roadway environment parameters, comprising a wind speed acquisition system, a fire bundle tube monitoring system, a high-level borehole gas drainage monitoring system, a corner gas concentration monitoring system, a PLC programmable logic controller (18), and a computer (4), characterized in that: The wind speed acquisition system is used to measure the average wind speed of the entire roadway cross-section at the positions of detection points A to D in real time. The wind speed acquisition system includes a transportation gateway (7), a return airway (8), the upper corner (2), and the lower corner (6). A mining roadway is provided between the mining face (1) and the goaf (3). The head and tail ends of the mining roadway are respectively connected to the transportation gateway (7) and the return airway (8). The connection between the transportation gateway (7) and the mining roadway is set as the lower corner (6). The side of the lower corner (6) close to the transportation gateway (7) is set as detection point A, and the side of the lower corner (6) close to the mining roadway is set as detection point B; the connection between the return airway (8) and the mining roadway is set as the upper corner (2). The side of the upper corner (2) close to the return airway (8) is set as detection point D, and the side of the upper corner (2) close to the mining roadway is set as detection point C. Ultrasonic wind speed measuring devices are respectively arranged at the positions of detection points A, B, C, and D. The fire bundle tube monitoring system is used to measure the gas concentration at different positions in the goaf in real time. The fire bundle tube monitoring system includes a number of upper corner fire bundle tubes (15), a number of lower corner fire bundle tubes (16), and a fire bundle tube analysis device (17). Along the length direction of the cross-section of the goaf (3) on the side close to the upper corner (2), a1, a2, a3... a are arranged in sequence. n Upper corner fire bundle tubes (15). Along the length direction of the cross-section of the goaf (3) on the side close to the lower corner (6), b1, b2, b3... b are arranged in sequence. n Lower corner fire bundle tubes (16). The fire bundle tube analysis device (17) is electrically connected to a number of upper corner fire bundle tubes (15) and lower corner fire bundle tubes (16) respectively. The high-level borehole gas drainage monitoring system is used to monitor the flow rate and CH4 concentration of high-level borehole gas drainage in real time. The high-level borehole gas drainage monitoring system includes a drainage pump (9) and a number of high-level borehole pipelines (5). The drainage pump (9) is located on one side of the transportation gateway (7). A mass flowmeter (10), a No. 1 concentration sensor (11), and a pressure sensor (12) are respectively arranged on the drainage pump (9); one end of the high-level borehole pipeline (5) extends into the goaf (3), and the other end of the high-level borehole pipeline (5) is connected to the drainage pump (9). The corner gas concentration monitoring system is used to monitor the gas concentration at the positions of the upper corner (2) and the lower corner (6) in real time. The corner gas concentration monitoring system includes a No. 2 concentration sensor (13) arranged at the position of the upper corner (2) and a No. 3 concentration sensor (14) arranged at the position of the lower corner (6). The PLC programmable logic controller (18) is set in the computer (4). The data signals collected by the wind speed acquisition system, the fire bundle tube monitoring system, the high-level borehole gas drainage monitoring system, and the corner gas concentration monitoring system are respectively connected to the signal input end of the PLC programmable logic controller (18) through signal lines. The PLC programmable logic controller (18) saves the data content to the computer (4), and processes the data output by the PLC through the software installed on the computer (4), and is then used for collaborative prediction of gas drainage and roadway environment parameters.

2. A method for collaborative prediction of gas drainage and roadway environmental parameters using the system described in claim 1, characterized in that, Including the following steps: S1. Use the gas drainage and roadway environment parameter collaborative prediction system to collect gas and fluid mechanics parameters of the mining face (1) at different drainage times, and build a scatter observation data set A1 (A 11 , A 12 , A 13 ); Wherein: Dataset A 11 (X 11 , t 11 , p 11 , u 11 , c 11 ) includes the position coordinates X of the drainage holes 11 and the CH4 drainage time t of the drainage pump (9) 11 , the drainage pressure p 11 , the drainage flow rate u 11 , the drainage concentration c 11 ; Dataset A 12 (X 12 , t 12 , p 12 , u 12 , c 12 ) includes the position coordinates X of different measurement points at the positions of the upper corner (2) and the lower corner (6) 12 and the drainage time t 12 , the drainage pressure p 12 , the roadway air velocity u 12 , the CH4 concentration c 12 ; Dataset A 13 (X 13 , t 13 , c 13 ) is the CH4 concentration c measured at different positions X of the goaf (3) at different times t 13 ; 13 The measured CH4 concentration c at different position coordinates X of the goaf (3) at different times t 13 ; When the geometric model is in two-dimensional state, the coordinates of X 11 , X 12 , X 13 are (x, y); when the geometric model is in three-dimensional state, the coordinates of X 11 , X 12 , X 13 are (x, y, z); S2. Use computational fluid dynamics software to simulate and calculate the flow characteristics of gas in a coal mine roadway, solve it through the set initial conditions and boundary conditions, and obtain the velocity u, pressure p, and concentration c at each position X(x i , x j ) at time t, and construct a numerical calculation result dataset A2(X, t, p, u, c); Wherein: The flow of gas in the goaf (3) satisfies the turbulent equation, the continuity equation, and the momentum conservation equation. The expression F1 is: In Equation (1), ρ is the gas density with the unit of kg / m 3 ; is a general variable; t is time; Γ is the generalized diffusion coefficient; S is the generalized source term; x i , x j are the coordinates of the three basic directions in the space coordinate system respectively, where i = 1, 2, or 3, and j = 1, 2, or 3; The gas diffusion equation F2 is: In Equation (2), c0 is the gas concentration; u is the gas velocity; S i is the discrete phase conversion rate; D is the diffusion coefficient of the gas; ▽ represents the Hamiltonian operator; Boundary conditions: In formula (3), Φ B is the boundary condition function; Γ i represents the i-th boundary; p i , c i , u i represent the pressure, concentration, and velocity of the i-th boundary condition; f(p i , c i , u i ) represents the function value of the i-th boundary condition; Initial conditions: In formula (4), Φ I is the initial value condition function; Ψ i represents the i-th initial value condition region; p i , c i , u i represent the pressure, concentration, and velocity of the i-th initial value condition respectively; g(p i , c i , u i ) represents the function value of the i-th initial value condition; S3. Data preprocessing: Combine the scatter observation data set A1 (A 11 , A 12 , A 13 ) constructed in step S1 and the numerical calculation result data set A2 (X, t, p, u, c) constructed in step S2, and perform preprocessing, including data cleaning and normalization operations, to form a data set A0, and divide it into a training set and a test set; S4. Deep learning model training: Use the multi-layer fully connected neural network algorithm to train the training set of the gas drainage system, and the outputs are p, u, and c. The activation function uses the smooth tanh function to ensure the existence and continuity of the derivative. Then, use the TensorFlow automatic differentiation technique and adopt the tape.gradient function to find the first-order and second-order derivatives of p, u, and c with respect to time and spatial coordinates. Construct the total loss function: L 总 = λ1L1 + λ PDE L PDE + λ BC L BC + λ IC L IC ; (5) In Equation (5), L1 is the loss function for dataset training, which is the MSE error; L PDE represents the residual loss of the partial differential equation; L BC represents the boundary condition loss; L IC represents the initial condition loss; λ1, λ PDE , λ BC and λ IC are all hyperparameters; ①Partial differential equation residual loss L PDE is the mean square error of the partial differential equation residual, as follows: L PDE = mean((F1 + F2) 2 ); (6) ②Boundary condition loss L BC : L BC = mean(Φ B 2 ); (7) ③ Initial condition loss L BC : L BC = mean(Φ I 2 ); (8) In Equations (6) to (8), mean is the mean function; S5. Adjust the weights and biases of the model through the gradient descent optimization algorithm to solve Arg minL 总 , until the prediction accuracy of the model converges, and obtain the physical parameters of the neural network link weight parameters and partial differential equations; S6. Use the trained model to evaluate the test set constructed in Step S5. The evaluation metrics include accuracy and precision. If the evaluation result is unqualified, repeat Steps S3 to S5 until the test set passes the evaluation, which is the optimized model. S7. Use the optimized model to jointly predict the gas concentration and gas drainage parameters at the upper corner position. S8. Gas over-limit warning: Adopt a single-index warning method, set the warning gas concentration to 1.0%, and when the predicted value of the gas concentration at the upper corner position reaches 95% of the limit value, give an early warning.

Images