A method and system for predicting coal drawing law of fully-mechanized caving mining based on graph neural network

Abstract

The application discloses a kind of based on graph neural network's fully mechanized top coal caving coal drawing law prediction method and system, it is related to mining engineering top coal drawing field, including: by discrete element software carries out the numerical simulation test of horizontal sectional fully mechanized top coal caving, establishes graph structure data set based on test result;Graph neural network is trained using graph structure data set, when loss function converges, the graph neural network that is trained is obtained;The node of graph is the position and physical information of particle in discrete element, edge is the contact state of each particle, global feature is the sectional height of horizontal sectional fully mechanized top coal face, coal seam thickness, coal seam dip, time step;Particle node position information after coal drawing is finished and coal-rock interface are obtained using trained graph neural network, and the shape of drawing body is obtained by inverting the coordinates of drawing particle.The present application can accurately predict top coal drawing body and coal-rock interface, and provides a fast and efficient research method for studying the complex coal drawing law of horizontal sectional fully mechanized top coal caving.

Description

Technical Field

[0001] This invention relates to the field of top coal caving mining technology, and more specifically to a method and system for predicting coal release patterns in fully mechanized top coal caving mining based on graph neural networks. Background Technology

[0002] Fully mechanized top coal caving mining technology is an effective method for mining thick coal seams. Among them, horizontal segmented fully mechanized top coal caving mining technology is one of the main methods for mining thick coal seams. Studying the top coal release law of top coal caving working faces is of great significance for guiding the actual operation process of top coal caving mining, optimizing coal release parameters, and improving the top coal recovery rate of the working face.

[0003] In horizontal segmented fully mechanized longwall mining, variations in coal seam occurrence conditions, thickness, dip angle, and segment height lead to changes in the coal release results, specifically manifested in differences in the development morphology of the top coal release body and the coal-rock interface. Currently, research methods for the top coal release body and coal-rock interface in top coal caving faces are limited, primarily relying on numerical simulation experiments, physical experiments, field measurements, or theoretical analysis. The specifics are as follows:

[0004] (1) Numerical simulation experiments can simulate the coal release process under various coal seam thicknesses and dip angles and accurately count the top coal release body and coal-rock interface morphology. However, due to the large thickness of the top coal in horizontal segment fully mechanized mining, the numerical simulation model will require a large number of particles, resulting in a long calculation time for the numerical simulation experiment. If we want to explore the development law of the top coal release body and coal-rock interface morphology under different segment heights, coal seam thicknesses and coal seam dip angles, a lot of calculation time is required.

[0005] (2) When using physical experiments to investigate the coal release law, the physical test platform is too complicated to build and the release body inversion process is difficult, resulting in poor repeatability of physical experiments under different coal seam thicknesses and coal seam dip angles, which is time-consuming and labor-intensive, and the data processing process is complicated.

[0006] (3) In the actual horizontal segmented fully mechanized mining face, it is basically unrealistic to conduct actual measurements of the coal-rock interface between the mined body and the coal. Although there are testing devices such as top coal migration trackers that can track and monitor the coal-rock interface on site, the setup of the testing instruments is complicated and affected by the complex construction environment, so the test results are often difficult to meet the research needs.

[0007] (4) When using theoretical analysis to study the coal release law, it is difficult to take into account the influence of complex boundary conditions and random flow of loose materials when establishing the theoretical equations. This results in poor generalization of the theoretical analysis results and makes it difficult to apply them well under different working face parameters (segment height, coal seam thickness, coal seam dip angle).

[0008] To overcome the shortcomings of the aforementioned research methods, prediction models based on deep learning algorithms have been widely applied across various industries as an efficient, fast, and accurate technique. Compared to numerical simulation experiments, algorithmic models can predict and output from any unknown input data based on training data, offering faster computation and more convenient results. Compared to physical experiments and field tests, algorithmic models have lower implementation costs and can minimize the consumption of human and material resources. Compared to traditional theoretical methods, algorithmic models can fully absorb and integrate multi-source data, demonstrating good applicability to various geological and boundary conditions.

[0009] Once the deep learning-based algorithm model is established, it eliminates the need for complex code to model different working conditions, allowing it to be directly used by most practitioners and greatly enhancing its application value. However, a review of existing literature and patents reveals a lack of research on algorithm models for predicting coal release patterns in horizontal segmented fully mechanized longwall mining based on graph neural networks. Therefore, proposing a graph neural network-based coal release pattern prediction scheme is of great significance for revealing the coal release patterns in horizontal segmented fully mechanized longwall mining, optimizing the coal release process, and improving resource recovery rates. This is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0010] In view of this, the present invention provides a method and system for predicting coal release patterns in fully mechanized longwall mining based on graph neural networks, which solves the problems existing in the background technology.

[0011] To achieve the above objectives, the present invention provides the following technical solution:

[0012] A method for predicting coal release patterns in fully mechanized longwall mining based on graph neural networks includes the following steps:

[0013] Numerical simulation experiments of horizontal segmented fully mechanized mining were conducted using discrete element method (DEM) software, and a graph-structured dataset was established based on the experimental results.

[0014] The graph neural network is trained using a graph structure dataset. When the loss function converges, the trained graph neural network is obtained. The nodes of the graph represent the position and physical information of the particles in the discrete element, the edges represent the contact state of each particle, and the global features are the segment height, coal seam thickness, coal seam dip angle, and time step of the horizontal segmented fully mechanized mining face.

[0015] The trained graph neural network is used to obtain the particle node position information and coal-rock interface after the coal discharge is completed. The shape of the discharged body is obtained by inverting the coordinates of the discharged particles.

[0016] Optionally, a numerical simulation experiment of horizontal segmented fully mechanized longwall mining can be conducted using discrete element method (DEM) software, specifically:

[0017] In PFC software, a basic numerical simulation model for horizontal segmented fully mechanized longwall mining was established. Coal release tests were conducted under different segment heights, working face lengths, coal seam dip angles, and coal release processes. Statistical analysis was performed on various information from the initial model, including... x Axis coordinates y Axial coordinates, particle radius, and category.

[0018] Optional, graph structure dataset The results consist of numerical simulation tests conducted under different combinations of segment height, coal seam thickness, and coal seam dip angle, specifically including:

[0019] Node characteristics: ,in, For the first i The ID of each particle. For the first i Each particle x coordinate, For the first i The y-coordinate of each particle, For the first i The type of particle, The particle size is denoted as .

[0020] Edge features: ,in, This is an edge index matrix composed of contact particle IDs. The edge feature matrix is ​​composed of contact force characteristics;

[0021] Global properties of a graph u This includes segment height, coal seam thickness, coal seam dip angle, and time step.

[0022] Optionally, during graph neural network training, the particle nodes and contact information are iteratively updated through L layers of message passing. The operation of each layer is as follows:

[0023] Message generation: For each edge Generate message ;in, For nodes i In the l Features of the layer For the edge In the l Features of the layer For message functions, For nodes j In the l Features of the layer Global property u In the l Features of the layer;

[0024] Message aggregation: based on For each node, aggregate the messages from its contact nodes; where, For aggregation operators, it represents attention weighting; For the edge In the l Messages from the layer For nodes i In the l The message aggregation result of the layer;

[0025] Node update: based on Update node i The characteristics; among which, For the update function, For nodes i In the l Features of +1 layer.

[0026] Optional, message function Choose either a multilayer perceptron (MLP) or a linear transform.

[0027] Optional, update function Select the gated recurrent unit (GRU).

[0028] Optionally, the loss function of the graph neural network can be obtained as follows:

[0029] The main loss function uses the mean squared error of location, and its expression is:

[0030]

[0031] In the formula: For mean square error loss, N For the sample size, Particles i The predicted value of the location, Particles i The true value of the location;

[0032] The expression for the morphological error of the emitted body is:

[0033]

[0034] In the formula: For segment height; h The distance from the center of the coal discharge port to the origin of the coal discharge port. , D The width of the coal discharge opening, The maximum transport angle of the granular material; This is the correction factor for the side shape of the base plate. This is the correction factor for the shape of the top plate side;

[0035] Let the left side of the equation be the equation of the released body, expressed as: ;in, This refers to the location information of the particle nodes. Coordinates of the coal outlet. S As a shape correction factor, the positional constraints of the particles released within the body or released from the particle body are written as: ;

[0036] For each coal-rock interface equation

[0037]

[0038] In the formula: Let be a function of the distance from a point on the equation to the coal outlet. For the first k Equation of coal-rock interface for secondary coal discharge. For the first k -1 Theoretical equation for the coal discharge body; Loss at the coal-rock interface;

[0039] The expression for the total error is:

[0040]

[0041] in, This is the regularity coefficient for the loss at the coal-rock interface.

[0042] A fully mechanized longwall mining coal release pattern prediction system based on graph neural network, which executes the fully mechanized longwall mining coal release pattern prediction method based on graph neural network as described above, includes a data acquisition module, a training module and a prediction module connected in sequence;

[0043] The data acquisition module is used to conduct numerical simulation experiments on horizontal segmented fully mechanized mining using discrete element method software, and to establish a graph-structured dataset based on the experimental results.

[0044] The training module is used to train the graph neural network using a graph structure dataset. When the loss function converges, the trained graph neural network is obtained. The nodes of the graph represent the position and physical information of the particles in the discrete element, the edges represent the contact state of each particle, and the global features are the segment height, coal seam thickness, coal seam dip angle, and time step of the horizontal segmented fully mechanized mining face.

[0045] The prediction module is used to obtain the particle node position information and coal-rock interface after the coal discharge is completed through a trained graph neural network, and to obtain the shape of the discharged body by inverting the coordinates of the discharged particles.

[0046] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method and system for predicting coal release patterns in fully mechanized longwall mining based on graph neural networks, which has the following beneficial effects:

[0047] (1) In order to solve the problems of long calculation time and difficulty in repeating experiments in numerical simulation test methods, the present invention constructs a graph neural network model, which can obtain the particle distribution characteristics after coal discharge by only inputting parameters such as the position information, contact information, segment height, coal seam thickness, and coal seam dip angle of the initial model, and then obtain the morphology of the coal-rock interface and the discharged body, making it faster and easier to obtain test results;

[0048] (2) In order to solve the problems of cumbersome process and data susceptibility to environmental influence in traditional physical experiments and field measurements, this invention is based on graph neural networks, which can realize the efficiency, convenience and safety of experimental research process;

[0049] (3) In order to solve the problem of poor applicability of theoretical research, this invention combines numerical simulation test results, physical test results and on-site measured numerical correction to realize the fusion of multi-source data and enhance the applicability of the final output results;

[0050] (4) In order to solve the problem that traditional graph neural network algorithms are difficult or impossible to apply in horizontal segmented fully mechanized mining, this invention proposes to introduce global features (segment height, coal seam thickness, coal seam dip angle) as well as position loss and coal-rock interface loss to improve the accuracy of the algorithm in the test results of horizontal segmented fully mechanized mining, and reduce the error caused by the inability of traditional graph neural networks to fully integrate and absorb the special boundary conditions and working parameters in the research of horizontal segmented fully mechanized mining. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0052] Figure 1 A flowchart of the method for predicting coal release patterns in fully mechanized longwall mining based on graph neural networks provided by the present invention;

[0053] Figure 2 This is a schematic diagram of the graph neural network used by the present invention to predict particle positions after coal discharge.

[0054] Figure 3 This invention provides a process for obtaining the top coal discharge body morphology through the output node information. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] This invention discloses a method for predicting coal release patterns in fully mechanized longwall mining based on graph neural networks, such as... Figure 1 As shown, it includes the following steps:

[0057] Numerical simulation experiments of horizontal segmented fully mechanized mining were conducted using discrete element method (DEM) software, and a graph-structured dataset was established based on the experimental results.

[0058] The graph neural network is trained using a graph structure dataset. When the loss function converges, the trained graph neural network is obtained. The nodes of the graph are the positions and physical information of the particles in the discrete elements, the edges are the contact states of each particle (including contacting particles and contact forces), and the global features are the segment height, coal seam thickness, coal seam dip angle, and time step of the horizontal segmented fully mechanized mining face.

[0059] The trained graph neural network is used to obtain the particle node position information and coal-rock interface after the coal discharge is completed. The morphology of the discharged body is obtained by inverting the coordinates of the discharged particles (the position of the particles in the initial model or the model before coal discharge).

[0060] according to Figure 1 As shown in the flowchart, this embodiment, based on numerical simulation test results, utilizes physical experiments, field measurement data, and reasonable physical constraints. Through the established graph neural network, it can predict coal release results under different segment heights, coal seam thicknesses, and coal seam dip angles, accurately obtaining the top coal release body and the coal-rock interface. This scheme boasts high computational efficiency, overcoming the time-consuming and cumbersome nature of numerical simulation experiments, physical experiments, and field tests. By integrating multi-scale physical constraints, it improves the accuracy of model prediction results, compensating for the poor applicability of theoretical methods. The model exhibits good generalization and applicability, providing a fast and efficient research method for studying the complex coal release laws in horizontal segmented fully mechanized longwall mining.

[0061] Furthermore, a numerical simulation experiment of horizontal segmented fully mechanized mining was conducted using discrete element method (DEM) software, specifically as follows:

[0062] In PFC software, a basic numerical simulation model for horizontal segmented fully mechanized longwall mining was established. Coal release tests were conducted under different segment heights, working face lengths, coal seam dip angles, and coal release processes. Statistical analysis was performed on various information from the initial model, including... x Axis coordinates yAxis coordinates, particle radius, and category. As shown in Table 1, taking a two-dimensional numerical simulation experiment as an example, if it is a three-dimensional experiment, then additional parameters are required. z Axis coordinates.

[0063] Table 1. Node information of the horizontal segmented fully mechanized mining diagram structure

[0064]

[0065] A graph-structured dataset is built based on the recorded data, containing edge messages as shown in Table 2 and global messages as shown in Table 3.

[0066] Table 2 Side Messages of Horizontal Segmented Fully Mechanized Mining Map Structure

[0067]

[0068] Table 3 Global Information on the Structure of Horizontal Segmented Fully Mechanized Mining Map

[0069]

[0070] Furthermore, graph structure datasets The results consist of numerical simulation tests conducted under different combinations of segment height, coal seam thickness, and coal seam dip angle, specifically including:

[0071] Node characteristics: ,in, For the first i The ID of each particle. For the first i Each particle x coordinate, For the first i The y-coordinate of each particle, For the first i The type of particle (0 for coal, 1 for gangue). The particle size is denoted as .

[0072] Edge features: ,in, This is an edge index matrix composed of contact particle IDs. The edge feature matrix is ​​composed of contact force characteristics;

[0073] Global properties of a graph u This includes segment height, coal seam thickness, coal seam dip angle, and time step.

[0074] Therefore, model input This means inputting the initial model with particle location information, contact information, and global information (segment height, coal seam thickness, and coal seam dip angle).

[0075] Furthermore, during graph neural network training, the particle nodes and contact information are iteratively updated through L-layer message passing. The operations of each layer are as follows:

[0076] Message generation: For each edge (from arrive ), generate message ;in, For nodes i (No. i (particle) in the first l Features of the layer For the edge In the l Features of the layer; For the message function, you can choose a multilayer perceptron (MLP), linear transform, etc. For nodes j In the l Features of the layer Global property u In the l Features of the layer;

[0077] Message aggregation: based on For each node (particle information), aggregate the messages of its contact nodes (particles); where, For aggregation operators, it represents attention weighting; For the edge In the l Messages from the layer For nodes i In the l The message aggregation result of the layer;

[0078] Node update: based on Update node i The characteristics; among which, For the update function, a gated recurrent unit (GRU) is selected; For nodes i In the l Features of +1 layer.

[0079] After L iterations of the above process, the predicted particle position is output:

[0080]

[0081] Output the position information of all particles after coal feeding is completed. Through the obtained particle position information After distinguishing the characteristics of coal and gangue, the coal-rock interface results can be obtained, such as... Figure 2 As shown. Then, the morphology of the emitted body can be inverted from the original model using the ID of the emitted particle, such as... Figure 3As shown.

[0082] Furthermore, the loss function of the graph neural network is obtained as follows:

[0083] The main loss function uses the mean squared error of location, and its expression is:

[0084]

[0085] In the formula: For mean square error loss, N For the sample size, Particles i The predicted value of the location, Particles i The true value of the location;

[0086] The expression for the morphological error of the emitted body is:

[0087]

[0088] In the formula: For segment height; h The distance from the center of the coal discharge port to the origin of the coal discharge port. , D The width of the coal discharge opening, The maximum transport angle of the granular material; This is the correction factor for the shape of the base plate side. This is the correction factor for the shape of the top plate side;

[0089] Let the left side of the equation be the equation of the released body, expressed as: ;in, This refers to the location information of the particle nodes. Coordinates of the coal outlet. S As a shape correction factor, the positional constraints of the particles released within the body or released from the particle body are written as: ;

[0090] For each coal-rock interface equation

[0091]

[0092] In the formula: Let be a function of the distance from a point on the equation to the coal outlet. For the first k Equation of coal-rock interface for secondary coal discharge. For the first k -1 Theoretical equation for the coal discharge body; Loss at the coal-rock interface;

[0093] The expression for the total error is:

[0094]

[0095] in, This is the regularity coefficient for the loss at the coal-rock interface.

[0096] and Figure 1 Corresponding to the method described above, this embodiment of the invention also provides a prediction system for coal release patterns in fully mechanized longwall mining based on graph neural networks, used for... Figure 1 The specific implementation of the method provided in this embodiment of the invention is a fully mechanized longwall mining coal release pattern prediction system based on graph neural network, which can be applied to computer terminals or various mobile devices. Specifically, it includes a data acquisition module, a training module and a prediction module connected in sequence.

[0097] The data acquisition module is used to conduct numerical simulation experiments on horizontal segmented fully mechanized mining using discrete element method software, and to establish a graph-structured dataset based on the experimental results.

[0098] The training module is used to train the graph neural network using a graph structure dataset. When the loss function converges, the trained graph neural network is obtained. The nodes of the graph represent the position and physical information of the particles in the discrete element, the edges represent the contact state of each particle, and the global features are the segment height, coal seam thickness, coal seam dip angle, and time step of the horizontal segmented fully mechanized mining face.

[0099] The prediction module is used to obtain the particle node position information and coal-rock interface after the coal discharge is completed through a trained graph neural network, and to obtain the shape of the discharged body by inverting the coordinates of the discharged particles.

[0100] In summary, this embodiment innovatively proposes a graph neural network-based coal release pattern prediction model for studying the coal release pattern in horizontal segmented fully mechanized longwall mining. The model's input includes node information (ID information, particle location information, physical information, category information), edge information (contact information between particles, contact force), and global attributes (segment height, coal seam thickness, coal seam dip angle, time step). The output includes particle location information after coal release, i.e., the coal-rock interface morphology. Notably, it can output the coal-rock interface morphology at various moments during continuous coal release and obtain the top coal release body morphology through inversion. Furthermore, the graph neural network loss function proposed in this embodiment is innovative. Compared to other graph neural networks that predict nodes or edges, this embodiment introduces global features (segment height, coal seam thickness, coal seam dip angle) as well as location loss, coal-rock interface loss, and physical constraints on the release body morphology to make the model results more accurate.

[0101] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0102] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A graph neural network-based fully mechanized caving coal drawing law prediction method, characterized in that, The method comprises the following steps: A numerical simulation test of horizontal sublevel fully mechanized mining is performed by using a discrete element software, and a graph structure dataset is established based on the test results; The graph neural network is trained using the graph structure dataset, and a trained graph neural network is obtained when the loss function converges; wherein, the nodes of the graph are the positions and physical information of the particles in the discrete element, the edges are the contact states of the particles, and the global features are the sublevel height of the horizontal sublevel fully mechanized working face, the coal seam thickness, the coal seam inclination, and the time step; in the loss function of the graph neural network, the position loss and the coal-rock interface loss are specifically introduced; The particle node position information and the coal-rock interface after the coal drawing is completed are obtained by using the trained graph neural network, and the drawing body shape is obtained by inverting the particle coordinates.

2. The method according to claim 1, wherein, The numerical simulation test of horizontal sublevel fully mechanized mining is performed by using a discrete element software, and the specific steps are as follows: In PFC software, the numerical simulation basic model of horizontal sublevel caving is established, and the caving test is carried out under the conditions of different sublevel height, working face length, coal seam inclination and caving technology. The information of the initial model is counted, including x axis coordinates, y axis coordinates, particle radius and category.

3. The method according to claim 1, wherein, Graph structure dataset The numerical simulation test results are composed of different segment heights, coal seam thicknesses, and coal seam inclination combinations, and specifically include: Node characteristics: ,in, For the first i The ID of each particle. For the first i Each particle x coordinate, For the first i The y-coordinate of each particle, For the first i The type of particle, The particle size is denoted as . edge features: wherein, is an edge index matrix consisting of contact particle ids, is an edge feature matrix consisting of contact force features; Global properties of the graph u Including segment height, coal seam thickness, coal seam dip angle and time step.

4. The graph neural network-based fully-mechanized top coal caving law prediction method according to claim 1, characterized in that, During the training of the graph neural network, the particle nodes and the contact information are updated through L-layer message passing iteration, and the operation of each layer is as follows: Message generation: For each edge Generate message ;in, For nodes i In the l Features of the layer For the edge In the l Features of the layer For message functions, For nodes j In the l Features of the layer Global property u In the l Characteristics of the layer; Message aggregation: based on aggregating messages of its contact nodes for each node; wherein, is an attention weight for the aggregation operator; is an edge In the first l layer, messages, is a node i In the first l layer, message aggregation results; Node update: according to update node i characteristics; wherein is an update function, is a node i the characteristics of the node at the first l +1 layer.

5. The graph neural network-based fully-mechanized top coal caving law prediction method according to claim 4, characterized in that, Message function Select a multi-layer perceptron, MLP, or a linear transformation.

6. The graph neural network-based fully-mechanized top coal caving law prediction method according to claim 4, characterized in that, Update function A gated recurrent unit (GRU) is selected.

7. The graph neural network-based fully-mechanized top coal caving law prediction method according to claim 1, characterized in that, The loss function of the graph neural network is obtained in the following manner: The main loss function uses the position mean square error, and the expression is as follows: In the formula: For mean square error loss, N For the sample size, Particles i The predicted value of the location, Particles i The true value of the location; The expression of the drawing body shape error is as follows: In the formula: is the subsection height; h is the distance from the center of the coal pass to the origin of the coal pass, , D is the coal pass width, is the maximum transport angle of the bulk material; is the floor side shape correction coefficient, is the roof side shape correction coefficient; Let the left side of the equation be the release body equation, expressed as ; wherein, is the particle node position information, is the release port coordinate, S is the shape correction coefficient, then the position constraint of the particle node in the release body or the released particle node is written as: ; The coal-rock interface equation of each time is as follows wherein: is a function of the distance from the point on the equation to the coal mouth, is the equation of the coal-rock interface for the k nth coal draw, is the equation of the coal-rock interface for the k -1 coal draw; is the coal-rock interface loss; The expression of the total error is as follows: wherein, is the regularization coefficient for the coal-rock interface loss.

8. A graph neural network-based fully-mechanized top coal caving coal drawing law prediction system, characterized in that, The method comprises the following steps: The method comprises the following steps: The data acquisition module is configured to perform a numerical simulation test of horizontal sublevel fully mechanized mining by using a discrete element software, and establish a graph structure dataset based on the test results; The training module is configured to train the graph neural network using the graph structure dataset, and obtain a trained graph neural network when the loss function converges; wherein, the nodes of the graph are the positions and physical information of the particles in the discrete element, the edges are the contact states of the particles, and the global features are the sublevel height of the horizontal sublevel fully mechanized working face, the coal seam thickness, the coal seam inclination, and the time step; in the loss function of the graph neural network, the position loss and the coal-rock interface loss are specifically introduced; The prediction module is configured to obtain the particle node position information and the coal-rock interface after the coal drawing is completed by using the trained graph neural network, and obtain the drawing body shape by inverting the particle coordinates.

Images