Knowledge data hybrid-driven coal mine underground roadway information modeling method

Abstract

The invention discloses a knowledge data hybrid-driven coal mine underground roadway information modeling method, which comprises the following steps of: establishing a roadway three-dimensional model based on a Workbench platform, setting boundary conditions, and setting spatio-temporal data output; processing roadway spatio-temporal data through a geostatistical interpolation method, and filling blank distribution of roadway spatial node wind speed and gas concentration information; selecting a proper partial differential equation as priori knowledge of the model according to physical information neural network characteristics and fluid flow characteristics; and introducing the partial differential equation into the loss function, and training a physical information neural network model conforming to physical constraints. The method can solve the problems of fluid mechanics differential equation solving and modeling in the ventilation process of the underground coal mine roadway.

Description

Technical Field

[0001] The present invention relates to the field of intelligent ventilation in underground coal mines, and in particular to a knowledge data hybrid driven coal mine tunnel information modeling method. Background Art

[0002] my country is the world's largest coal producer and consumer, with rich coal reserves, ranking first in the world. Coal has long occupied a core position in my country's energy structure. It not only supports the country's industrialization process, but also makes great contributions to social and economic development. However, most of my country's coal resources are distributed underground, and the geological conditions are complex and changeable. With the continuous mining of mines, the environment under coal mines is becoming increasingly complex. During deep mining, the geological structure inside the mine may become unstable, especially in areas with developed faults and fissures, where the risk of accumulation of harmful gases such as gas is extremely high. By timely grasping the environmental information of mine tunnels, potential safety risks can be effectively warned and the level of mine safety production can be improved. However, traditional tunnel monitoring collects data through sensors at fixed locations, and the coverage is limited and cannot characterize the environmental information of the entire tunnel; using traditional deep neural networks to establish tunnel information models requires a large amount of data, and the lack of physical knowledge leads to poor model generalization ability.

[0003] With the help of geostatistical interpolation methods, the wind speed, gas concentration, pressure and other parameters in the tunnel are interpolated to obtain the distribution of each spatial position. This method can not only effectively fill the blank areas in monitoring, but also provide important data support for comprehensive evaluation and monitoring. Through reconstruction technology, the environmental information in the tunnel can be more accurately grasped, and potential dangerous areas can be warned in advance, thereby improving the scientificity and effectiveness of mine safety management. In addition, the introduction of physical knowledge constraints and the establishment of a tunnel information model in combination with physical information neural networks can greatly reduce the demand for training data and enhance the generalization ability of the model. This modeling technique helps to improve the level of intelligence of underground ventilation in coal mines, fill in the blank areas in monitoring, and warn potential dangerous areas in advance, further promoting the development of smart mines. Summary of the invention

[0004] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a knowledge data hybrid driven coal mine underground tunnel information modeling method, which can solve the problems of solving and modeling fluid mechanics differential equations in the ventilation process of coal mine underground tunnels.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] The present invention provides a knowledge data hybrid driven coal mine tunnel information modeling method, comprising the following steps:

[0007] Step 1: Based on the Workbench platform, establish a three-dimensional model of the tunnel, set boundary conditions, and set the spatiotemporal data output to ASCII format;

[0008] Step 2: Process the tunnel spatiotemporal data by using geostatistical interpolation methods to fill in the blank distribution of wind speed and gas concentration information at tunnel spatial nodes;

[0009] Step 3, according to the physical information neural network characteristics and fluid flow characteristics, select appropriate partial differential equations as prior knowledge of the model;

[0010] Step 4: Introduce partial differential equations into the loss function and train a physical information neural network model that conforms to physical constraints.

[0011] Preferably, step 1 specifically includes the following contents:

[0012] S11. Establish a three-dimensional model of the tunnel according to the actual tunnel structure and size;

[0013] S12. Setting the boundary conditions in the tunnel according to the coal mining and cutting method, air inlet, air return port and wind direction of the underground working face of the coal mine;

[0014] S13, selecting a suitable simulation convection model by calculating the Reynolds number Re;

[0015] S14, adding gas as a diffusion source to the components, and setting a position on the coal mining and cutting wall of the working face in the coal mine;

[0016] S15. According to the working process of the coal mining machine, the speed and position of the diffusion source are set, and a UDF expression is written;

[0017] S16. Select the tunnel time and space information, including pressure P, wind speed components U, V, W in the X, Y, Z directions, and gas concentration C as feature outputs to obtain the required data.

[0018] Preferably, step 2 specifically includes the following contents:

[0019] S21, converting the ASCII data format obtained in step 1 into an Excel data format;

[0020] S22, according to the spatial node information of the three-dimensional model of the coal mine tunnel space, the inverse distance weighted interpolation method is used for interpolation calculation; at this time, the positions of each spatial node in the tunnel are (x i ,y i ,z i ), node information is f i , i is the number of surrounding nodes that need to be calculated to determine the target node, the spatial position of the target node is (x, y, z), and the distance between spatial nodes is The target node information f(x,y,z) that needs to be solved is:

[0021]

[0022] S23, according to the calculation formula in step S22, determine the number of surrounding nodes calculated by the target node, and calculate the information of the target node;

[0023] S24. Store the processed data into a new Excel file so that the physical information neural network model can be trained in the subsequent steps.

[0024] Preferably, step 3 specifically includes the following contents:

[0025] S31. Select a classical physical information neural network and analyze its network structure;

[0026] S32. According to the state of the fluid in step 1 being in turbulence, the fluid mechanics Navier-Stokes equation is selected, where c is the gas concentration, x, y, z are the three-dimensional space coordinate directions, ρ is the fluid density, u, v, w are the components of the wind speed in the x, y, z directions, D is the diffusion coefficient, p is the fluid pressure, μ is the dynamic viscosity, and the equation is expressed as follows:

[0027] Convection Diffusion:

[0028]

[0029] S33. According to the gas diffusion source in step 2, the Navier-Stokes equation is improved and perfected by adding the diffusion source term S c , S diff,x , S diff,y , S diff,z is the component of the diffusion source term in the x, y, and z directions, g x , g y , g z is the component of gravity acceleration in the x, y, and z directions. The expression is:

[0030] Convection Diffusion:

[0031]

[0032] S34. Adjust the position of the data columns according to the partial differential equation to provide conditions for model training.

[0033] Preferably, step 4 specifically includes the following contents:

[0034] S41. Convert the Navier-Stokes equations into automatic differentiation form;

[0035] S42, input the data (x, y, z, t) calculated in step 2 into the deep neural network;

[0036] S43, taking the pressure P, wind speed components U, V, W, and gas concentration C data of the number of columns adjusted in step 3 as target values ​​for training;

[0037] S44, bringing the output obtained in step S42 into the partial differential equation as a numerical solution to perform automatic differential calculation;

[0038] S45, introducing the automatic differentiation result into the loss function, merging it with the current loss function, and using them together as a new loss function;

[0039] S46. During the model training process, the gradient is calculated through the latest loss function, and the network parameters are adjusted;

[0040] S47, continuously iterating during the training process until the loss value of the model on the training set and the validation set reaches a preset standard, thereby obtaining an accurate coal mine underground tunnel information model;

[0041] S48. Verify the model by inputting spatiotemporal information outside the training set.

[0042] The beneficial effects of the present invention are:

[0043] 1. This method simulates the tunnel information under the Workbench platform and processes the data through the geostatistical interpolation method to improve the information of the missing space of the tunnel, providing a more comprehensive and accurate data basis for the physical information neural network modeling, making the model more accurate.

[0044] 2. Combination of data-driven and knowledge-driven: By introducing fluid mechanics differential equations, the constraints of real-world fluids are taken into account to ensure consistency with real physical conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 It is a flow chart of a method for modeling coal mine underground tunnel information driven by a knowledge data hybrid method provided by an embodiment of the present invention;

[0047] Figure 2 is a three-dimensional structural diagram of a lane provided by an embodiment of the present invention;

[0048] Figure 3 is a structural principle diagram of a physical information neural network provided by an embodiment of the present invention;

[0049] Figure 4 is a graph of the loss function of the training process provided by an embodiment of the present invention;

[0050] Figure 5 is a wind speed prediction verification curve provided by an embodiment of the present invention;

[0051] Figure 6 It is a gas concentration prediction verification curve provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] See also Figure 1 This embodiment provides a knowledge data hybrid driven coal mine tunnel information modeling method, comprising the following steps:

[0054] Step 1 is implemented using the Workbench platform, and the data is output in ASCII format, including the following:

[0055] S11, such as Figure 2 As shown, a three-dimensional model is established based on the actual tunnel structure, including air inlet and return air outlet, gas diffusion position, tunnel topography, tunnel structure and tunnel size;

[0056] S12. Setting boundary conditions in the tunnel according to the coal mining and cutting method, air inlet and return air outlet, and wind direction of the underground working face of the coal mine;

[0057] S13, selecting a suitable simulation model by calculating the Reynolds number Re;

[0058] S14, adding gas as a diffusion source to the components, and setting a position on the coal mining and cutting wall of the working face in the coal mine;

[0059] S15. According to the working process of the coal mining machine, the speed and position of the diffusion source are set, and a UDF expression is written;

[0060] S16, select the tunnel time and space information, including pressure P, wind speed components U, V, W in the X, Y, Z directions,

[0061] The gas concentration C is used as the characteristic output to obtain the required data;

[0062] Step 2: Data processing based on the inverse distance weighted interpolation method; the details are as follows:

[0063] S21, converting the ASCII data format obtained in step 1 into an Excel data format;

[0064] S22, according to the spatial node information of the three-dimensional model of the coal mine tunnel space, the inverse distance weighted interpolation method (IDW) is used for interpolation calculation. At this time, the positions of each spatial node in the tunnel are (x i ,y i ,z i ), the spatial position of the target node is (x, y, z), and the node information is f i , i is the number of surrounding nodes that need to be calculated to determine the target node, and the distance between spatial nodes is The target node information f(x,y,z) that needs to be solved is:

[0065]

[0066] S23, according to the calculation formula in S22, determine the number of surrounding nodes calculated by the target node, and calculate the information of the target node;

[0067] S24, storing the processed data in a new Excel file so as to train the physical information neural network model in the subsequent steps;

[0068] Step 3, combining the physical information neural network characteristics and fluid flow characteristics, select the Navier-Stokes equation as prior knowledge, the specific content is as follows:

[0069] S31. Select a classical physical information neural network, whose network structure is as follows Figure 3 As shown;

[0070] S32. According to the state of the fluid in step 1 being in turbulence, the fluid mechanics Navier-Stokes equation is selected, where c is the gas concentration, x, y, z are the three-dimensional space coordinate directions, ρ is the fluid density, u, v, w are the components of the wind speed in the x, y, z directions, D is the diffusion coefficient, p is the fluid pressure, μ is the dynamic viscosity, and the equation is expressed as follows:

[0071] Convection Diffusion:

[0072]

[0073] S33. According to the gas diffusion source in step 2, the Navier-Stokes equation is improved by adding the diffusion source term Sc , S diff,x , S diff,y , S diff,z is the component of the diffusion source term in the x, y, and z directions, g x , g y , g z is the component of gravity acceleration in the x, y, and z directions. The expression is:

[0074] Convection Diffusion:

[0075]

[0076] S34. Adjust the position of the data columns according to the partial differential equation to provide convenient conditions for model training;

[0077] Step 4: Use the pytorch framework to train the model on the PyCharm platform. The details are as follows:

[0078] S41. Convert the Navier-Stokes equations into automatic differentiation form through programming;

[0079] S42, input the data (x, y, z, t) calculated in step 2 into the deep neural network;

[0080] S43, taking the pressure P, wind speed components U, V, W, and gas concentration C data of the number of columns adjusted in step 3 as target values ​​for training;

[0081] S44, bringing the output obtained in S42 into the partial differential equation as a numerical solution to perform automatic differential calculation;

[0082] S45, introduce the automatic differentiation result into the loss function, merge it with the current loss function, and use them together as a new loss function. The loss function curve during the training process is as follows: Figure 4 As shown;

[0083] S46. During the model training process, the gradient is calculated through the latest loss function, and the network parameters are adjusted;

[0084] S47, continuously iterating during the training process until the loss value of the model on the training set and the validation set reaches a preset standard, thereby obtaining an accurate coal mine underground tunnel information model;

[0085] S48. The model is verified by inputting spatiotemporal information outside the training set. The wind speed prediction verification curve is as follows: Figure 5 As shown in Figure 2, the gas concentration prediction verification curve is as follows: Figure 6 As shown, the black solid line is the real data, and the black dotted line is the predicted output data.

[0086] This method simulates the tunnel information under the Workbench platform, processes the data through geostatistical interpolation methods, improves the information of the missing space of the tunnel, and provides a more comprehensive and accurate data basis for physical information neural network modeling, making the model more accurate. It adopts a combination of data-driven and knowledge-driven: by introducing fluid mechanics differential equations, the constraints of real-world fluids are taken into account to ensure consistency with real physical conditions.

[0087] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A coal mine underground tunnel information modeling method driven by knowledge data hybrid, characterized in that: The following steps are involved: Step 1: Based on the Workbench platform, establish a three-dimensional model of the tunnel, set boundary conditions, and set the spatiotemporal data output to ASCII format; Step 2: Process the tunnel spatiotemporal data by using geostatistical interpolation methods to fill in the blank distribution of wind speed and gas concentration information at tunnel spatial nodes; Step 3, according to the physical information neural network characteristics and fluid flow characteristics, select appropriate partial differential equations as prior knowledge of the model; Step 4: Introduce partial differential equations into the loss function and train a physical information neural network model that conforms to physical constraints.

2. A coal mine underground tunnel information modeling method driven by knowledge data hybrid as claimed in claim 1, characterized in that: Step 1 specifically includes the following: S11. Establish a three-dimensional model of the tunnel according to the actual tunnel structure and size; S12. Setting the boundary conditions in the tunnel according to the coal mining and cutting method, air inlet, air return port and wind direction of the underground working face of the coal mine; S13, selecting a suitable simulation convection model by calculating the Reynolds number Re; S14, adding gas as a diffusion source to the components, and setting a position on the coal mining and cutting wall of the working face in the coal mine; S15. According to the working process of the coal mining machine, the speed and position of the diffusion source are set, and a UDF expression is written; S16. Select the tunnel time and space information, including pressure P, wind speed components U, V, W in the X, Y, Z directions, and gas concentration C as feature outputs to obtain the required data.

3. A coal mine underground tunnel information modeling method driven by knowledge data hybrid as claimed in claim 2, characterized in that: Step 2 specifically includes the following: S21, converting the ASCII data format obtained in step 1 into an Excel data format; S22, according to the spatial node information of the three-dimensional model of the coal mine tunnel space, the inverse distance weighted interpolation method is used for interpolation calculation; at this time, the positions of each spatial node in the tunnel are (x i ,y i ,z i ), the spatial position of the target node is (x, y, z), and the node information is f i , i is the number of surrounding nodes that need to be calculated to determine the target node, and the distance between spatial nodes is The target node information f(x,y,z) that needs to be solved is: S23, according to the calculation formula in step S22, determine the number of surrounding nodes calculated by the target node, and calculate the information of the target node; S24. Store the processed data into a new Excel file so that the physical information neural network model can be trained in the subsequent steps.

4. A method for modeling underground mine tunnel information driven by knowledge data hybrid as claimed in claim 3, characterized in that: Step 3 specifically includes the following: S31. Select a classical physical information neural network and analyze its network structure; S32. According to the state of the fluid in step 1 being in turbulence, the fluid mechanics Navier-Stokes equation is selected, where c is the gas concentration, x, y, z are the three-dimensional space coordinate directions, ρ is the fluid density, u, v, w are the components of the wind speed in the x, y, z directions, D is the diffusion coefficient, p is the fluid pressure, μ is the dynamic viscosity, and the equation is expressed as follows: Convection Diffusion: S33. According to the gas diffusion source in step 2, the Navier-Stokes equation is improved and perfected by adding the diffusion source term S c , S diff,x , S diff,y , S diff,z are the components of the diffusion source term in the x, y, and z directions, g x , g y , g z is the component of gravity acceleration in the x, y, and z directions. The expression is: Convection Diffusion: S34. Adjust the position of the data columns according to the partial differential equation to provide conditions for model training.

5. A method for modeling coal mine underground tunnel information driven by knowledge data hybrid as claimed in claim 4, characterized in that: Step 4 specifically includes the following: S41. Convert the Navier-Stokes equations into automatic differentiation form; S42, input the data (x, y, z, t) calculated in step 2 into the deep neural network; S43, taking the pressure P, wind speed components U, V, W, and gas concentration C data of the number of columns adjusted in step 3 as target values ​​for training; S44, bringing the output obtained in step S42 into the partial differential equation as a numerical solution to perform automatic differential calculation; S45, introducing the automatic differentiation result into the loss function, merging it with the current loss function, and using them together as a new loss function; S46. During the model training process, the gradient is calculated through the latest loss function, and the network parameters are adjusted; S47, continuously iterating during the training process until the loss value of the model on the training set and the validation set reaches a preset standard, thereby obtaining an accurate coal mine underground tunnel information model; S48. Verify the model by inputting spatiotemporal information outside the training set.

Images