A method for predicting coal spontaneous combustion temperature based on graph convolutional neural network

Abstract

The present invention discloses a method for predicting the spontaneous combustion temperature of coal based on a graph convolutional neural network, which relates to the technical field of coal mine fire prediction, including obtaining information about a variety of gases generated when the current coal spontaneously combusts in the air; using a temperature prediction model to predict the temperature using the information about the various gases, and obtaining the current spontaneous combustion temperature of the coal. The temperature prediction model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, and a fully connected layer connected in sequence; the input layer is used to embed the information about the various gases into a directed connection graph. The present invention improves the prediction accuracy of the spontaneous combustion temperature of coal.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine fire prediction, and in particular to, but not limited to, a method for predicting coal spontaneous combustion temperature based on a graph convolutional neural network, an electronic device and a storage medium. Background Art

[0002] Coal is a common fossil fuel, but under certain conditions, such as during storage or transportation, coal may spontaneously combust, producing fires and harmful gases. Spontaneous combustion is one of the main disasters that affect the safe and orderly production of coal mines. It not only causes waste of coal resources and manpower and material resources, but also lays the hidden dangers for inducing gas explosions and coal dust explosions, which in turn causes immeasurable casualties and property losses.

[0003] In the related art, the spontaneous combustion temperature of coal is predicted by a single gas or an index gas. In the actual environment, the concentration of a single index gas is greatly affected by the environment, and the relationship between different gases is nonlinear, making it difficult to accurately predict the coal temperature.

[0004] Therefore, how to quickly and accurately predict the temperature of coal spontaneous combustion has become an urgent problem to be solved. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a method for predicting coal spontaneous combustion temperature based on a graph convolutional neural network, which at least solves the problem that the related technology has low accuracy in predicting the temperature during coal spontaneous combustion.

[0006] According to a first aspect of an embodiment of the present invention, a method for predicting coal spontaneous combustion temperature based on a graph convolutional neural network is provided, comprising:

[0007] Obtain information on various gases produced when coal spontaneously combusts in the air;

[0008] The temperature prediction model uses the multiple gas information to perform temperature prediction to obtain the current coal self-ignition temperature. The temperature prediction model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer and a fully connected layer connected in sequence; the input layer is used to embed the multiple gas information into a directed connection graph.

[0009] The solution provided by the embodiment of the present invention obtains information about various gases generated when the current coal spontaneously combusts in the air; the temperature prediction model uses the information about various gases to perform temperature prediction, and obtains the spontaneous combustion temperature of the current coal. The temperature prediction model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, and a full connection connected in sequence; the input layer is used to embed the information about various gases into a directed connection graph. In this method, a directed connection graph is constructed by processing the information about various gases using the input layer, and the first convolutional layer, the second convolutional layer, and the third convolutional layer in the temperature prediction model perform convolution operations on the constructed directed connection graph, effectively learning the local neighborhood information of the node, and combining it with the characteristics of the node itself, thereby extracting a richer and more useful feature representation. These features are used to reveal the complex nonlinear relationship between gas concentration and the spontaneous combustion temperature of coal, thereby achieving accurate prediction of the spontaneous combustion temperature of coal. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0011] Figure 1 A schematic flow chart of a method for predicting spontaneous combustion temperature of coal based on a graph convolutional neural network provided in an embodiment of the present invention;

[0012] Figure 2 A schematic diagram of the effect of a directed connection graph provided by an embodiment of the present invention;

[0013] Figure 3 A schematic diagram of a method for predicting the spontaneous combustion temperature of coal provided by an embodiment of the present invention;

[0014] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0016] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0017] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present invention described here can be implemented in an order other than that illustrated or described here.

[0018] Those skilled in the art will appreciate that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the field to which the embodiments of the present invention belong. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0019] Figure 1 A flow chart of a method for predicting spontaneous combustion temperature of coal based on a graph convolutional neural network provided in an embodiment of the present invention is provided. The method for predicting spontaneous combustion temperature of coal provided in an embodiment of the present invention can be executed by an electronic device, and the electronic device can be a computer, a server, etc.

[0020] like Figure 1 As shown in the figure, the coal spontaneous combustion temperature prediction method based on graph convolutional neural network includes:

[0021] S101. Obtain information on various gases generated when coal spontaneously combusts in the air.

[0022] In the embodiments of the present invention, "coal spontaneous combustion" refers to the phenomenon that coal undergoes physical and chemical reactions under long-term contact with oxygen in the air, resulting in a rise in temperature and eventually causing spontaneous combustion. In addition, coal will produce a variety of gases during the spontaneous combustion process, and the various gas information includes but is not limited to the various gases. , , , , and As well as information such as the corresponding gas concentration, production rate, and oxygen consumption rate.

[0023] S102, using the temperature prediction model to use the gas information to perform temperature prediction, and obtain the current spontaneous combustion temperature of the coal. The temperature prediction model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer and a fully connected layer connected in sequence; the input layer is used to embed multiple gas information into a directed connection graph.

[0024] In an embodiment of the present invention, the temperature prediction model includes an input layer, three convolutional layers and a fully connected layer. The input layer is directly connected to the first convolutional layer, the first convolutional layer is directly connected to the second convolutional layer, the second convolutional layer is directly connected to the third convolutional layer, and the third convolutional layer is directly connected to the fully connected layer. An activation function can be set between the input layer and the first convolutional layer, an activation function can be set between the first convolutional layer and the second convolutional layer, and an activation function can be set between the second convolutional layer and the third convolutional layer. The temperature prediction model can be a graph convolutional neural network. Multiple gas information is input into the trained temperature prediction model to obtain the current spontaneous combustion temperature of the coal.

[0025] In an embodiment of the present invention, model training is required before using the temperature prediction model. During the training stage, the mean square error (MSE) loss (expressed as LMSE) is used to calculate the minimum error between the predicted temperature TP and the actual temperature TT. The Adam optimizer is used to update the weight parameters in the model by minimizing the mean square error loss until the training conditions are met to obtain a trained temperature prediction model.

[0026] In an embodiment of the present invention, after the gas information is input into the input layer of the temperature prediction model, the gas information is embedded in the input layer to obtain a directed connection graph. The directed connection graph is composed of multiple nodes and directed edges, wherein the node can represent each gas and an attribute value associated with each gas, and the attribute value includes gas concentration, gas oxygen consumption rate, gas generation rate, gas generation rate / gas consumption rate and gas ratio, and the gas consumption rate includes gas oxygen consumption rate and gas generation rate. The directed edges represent the generation relationship between each node. For example, in the process of coal spontaneous combustion, the concentration change of various gases is affected by the oxygen concentration. Therefore, in the directed connection graph The concentration indicates the concentration of other gases, and the ratios between different gases are indicated by the concentrations of each gas. Production rate, The generation rate is obtained by , , The Graham coefficient is calculated from the oxygen consumption rate, Production rate , Production rate Income, so Concentration refers to the oxygen consumption rate, Concentration direction , Concentration direction , and different Graham coefficients are pointed to by their respective gas change rates.

[0027] In an embodiment of the present invention, as shown in the following Table 1, Table 1 shows the nodes of a directed connection graph and the meaning of each node.

[0028] Table 1 Nodes of a directed connection graph and the meaning of each node

[0029] ;

[0030] For example, Figure 2 As shown, Figure 2 A schematic diagram of the effect of a directed connection graph provided by an embodiment of the present invention. Figure 2 The directed connection graph in contains 17 nodes in total, and each node is connected by a directed edge.

[0031] It can be understood that in the embodiment of the present invention, the information of various gases generated when the current coal spontaneously combusts in the air is obtained; the temperature prediction model uses the various gas information to predict the temperature, and obtains the current coal spontaneous combustion temperature. The temperature prediction model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer and a fully connected layer. The first convolutional layer is connected after the input layer, the second convolutional layer is connected after the first convolutional layer, the third convolutional layer is connected after the second convolutional layer, and the fully connected layer is connected after the third convolutional layer; the input layer is used to embed the various gas information into a directed connection graph. In this method, by processing the various gas information using the input layer to construct a directed connection graph, and performing convolution operations on the first convolutional layer, the second convolutional layer and the third convolutional layer in the temperature prediction model and the constructed directed connection graph, the local neighborhood information of the node is effectively learned, and it is combined with the characteristics of the node itself, so as to extract a richer and more useful feature representation. These features are used to reveal the complex nonlinear relationship between gas concentration and coal spontaneous combustion temperature, thereby realizing accurate prediction of coal spontaneous combustion temperature.

[0032] In some embodiments of the present invention, S102 can be implemented through S1021 to S1023, which is specifically described by the following steps.

[0033] S1021. Input various gas information into the input layer to obtain a directed connection graph.

[0034] S1022. Perform convolution operations on the directed connection graph in sequence based on the first filter, the second filter, and the third filter to obtain a target feature matrix.

[0035] In some embodiments of the present invention, after inputting the multiple gas information into the input layer, the input layer embeds the multiple gas information into a directed graph structure to generate embedded information, and the feature dimension of each node in the directed connection graph is 1. The first convolution layer includes a first filter, the second convolution layer includes a second filter, and the third convolution layer includes a third filter, and each filter is used for feature extraction. The number of first filters is equal to the number of third filters, and the number of second filters can be greater than the number of first filters. After the first filter performs a convolution operation on the directed connection graph, a first feature matrix is ​​obtained, and then the second filter performs a convolution operation on the first feature matrix to obtain a second feature matrix, and finally the third filter performs a convolution operation on the second feature matrix to obtain a target feature matrix.

[0036] Exemplarily, the number of the first filters and the third filters may be 16, and the number of the second filters may be 32.

[0037] S1023. Fit the target feature matrix through the fully connected layer to obtain the auto-ignition temperature.

[0038] In some embodiments of the present invention, after receiving the target feature matrix, the fully connected layer performs linear transformation and applies activation functions and other operations to fit the target feature matrix to obtain the current spontaneous combustion temperature of the coal.

[0039] like Figure 3 As shown, Figure 3 The present invention provides an architecture diagram of a method for predicting coal spontaneous combustion temperature based on a graph convolutional neural network. Figure 3 In the graph convolutional neural network, the temperature prediction model is a graph convolutional neural network, including an input layer, convolutional layer 1, convolutional layer 2, convolutional layer 3 and a fully connected layer, wherein an activation function (ReLU) is set between the input layer and convolutional layer 1, an activation function (ReLU) is set between convolutional layer 1 and convolutional layer 2, and an activation function (ReLU) is set between convolutional layer 2 and convolutional layer 3. The collected multiple gas information (i.e., input features) is input into the input layer of the graph convolutional neural network, wherein the multiple gas information may include oxygen, carbon monoxide, carbon dioxide, etc., and each gas can be associated with an attribute value. The input layer embeds the input multiple gas information into a directed graph to obtain a directed connection graph. The directed connection Figure 1 There are 17 nodes in total, and the feature dimension of each node is 1. The directed connection graph is processed by the activation function and then input into the convolution layer 1. After the node information aggregation and other operations, the convolution layer 1 maps the 1-dimensional features of each node to the 16-dimensional space to obtain the first feature matrix; then the first feature matrix is ​​processed by the activation function ( , X is the node) is processed and input into the convolution layer 2. After node information aggregation and other operations, the 16-dimensional features of each node are mapped to the 32-dimensional space to obtain the second feature matrix; finally, the second feature matrix is ​​processed by the activation function and input into the convolution layer 3. After node information aggregation and other operations, the 32-dimensional features of each node are compressed into the 16-dimensional space to obtain the target feature matrix (17 16). The target feature matrix is ​​flattened and input into the fully connected layer. The fully connected layer outputs the predicted temperature of coal spontaneous combustion based on the target feature matrix. is the self-connected adjacency matrix of the directed connected graph, Matrix and Matrix are the degree matrix and the inverse matrix of the degree matrix corresponding to the self-connected adjacency matrix of the directed connected graph, respectively. is the weight matrix of convolutional layer 1, Matrix is the node matrix of convolutional layer 1, Matrix is the node matrix of convolutional layer 2, Feature vector H is the feature vector extracted by convolutional layer 1, and so on.

[0040] It is understandable that in some embodiments of the present invention, multiple gas information is input into the input layer to obtain a directed connection graph, and the directed connection graph is sequentially convolved using the first filter, the second filter, and the third filter to obtain a target feature matrix, and the target feature matrix is ​​fitted through the fully connected layer to obtain the auto-ignition temperature. This method uses multiple gas information to establish a graph structure of nodes and edges, processes the input node features through the convolution layer of the temperature prediction model, maps the features of each node to a high-dimensional space, enriches the features of each node, generates a new feature representation for each node, helps the model aggregate the features of each node, and the fully connected layer integrates the node features processed by the convolution layer and outputs the auto-ignition temperature, thereby improving the accuracy of the auto-ignition temperature.

[0041] In some embodiments of the present invention, S1022 may be implemented through S201 to S203, which is specifically described by the following steps.

[0042] S201. Perform a convolution operation on the directed connection graph using a first filter to obtain a first feature matrix.

[0043] In some embodiments of the present invention, the feature dimension of each node of the directed connection graph is 1, the number of first filters is multiple, and the relevant information of each node is input into multiple filters for convolution operation, so that the feature dimension of each node is increased and new features are generated through each first filter, thereby achieving the purpose of dimensionality increase and feature extraction, and obtaining the first feature matrix.

[0044] Exemplarily, the number of first filters is 16, and the directed connection graph is input into the first convolutional layer. The feature dimension of each node is 1, and the features of each node are respectively passed through 16 filters to generate new features to form a first feature matrix, thereby mapping the 1-dimensional features of each node to a 16-dimensional space and extracting high-dimensional representations of node features, thereby capturing the complex interactions and dependencies between nodes and enhancing the learning ability of the model.

[0045] S202: Use a second filter to perform a convolution operation on the first feature matrix to obtain a second feature matrix.

[0046] In some embodiments of the present invention, there are multiple second filters, and the first feature matrix is ​​input into the second filter. In the first feature matrix, the feature dimension of each node is 16. Each node with a feature dimension of 16 is input into multiple second filters for convolution operations, so that the feature dimension of each node is increased and new features are generated through each second filter, thereby achieving the purpose of dimensionality increase and feature extraction, and obtaining the second feature matrix.

[0047] Exemplarily, the number of second filters is 32, and the first feature matrix is ​​input into the second convolutional layer. The feature dimension of each node is 16, and each node generates new features through 32 second filters to form a second feature matrix, thereby mapping the 16-dimensional features of each node to a 32-dimensional space, thereby further enriching the feature information of each node.

[0048] S203: Use the third filter to perform a convolution operation on the second feature matrix to obtain a target feature matrix.

[0049] In some embodiments of the present invention, the number of the third filters is multiple, and the number of the third filters may be equal to the number of the first filters. The second feature matrix is ​​input into the third filter, in which the feature dimension of each node is 32, and each node with a feature dimension of 32 is input into multiple third filters for convolution operation, so that the feature dimension of each node is increased and new features are generated through each third filter, thereby obtaining a target feature matrix.

[0050] Exemplarily, the number of third filters is 16, and the second feature matrix is ​​input into the third convolutional layer. The feature dimension of each node is 32. Each node generates new features through 16 third filters to form a target feature matrix, thereby compressing the 32-dimensional features of each node into a 16-dimensional space. The third convolutional layer is connected to the fully connected layer to reduce the computational complexity of the fully connected layer and better aggregate the features of each node for temperature prediction.

[0051] It can be understood that in some embodiments of the present invention, the first convolution layer maps the feature dimension of each node to, for example, a 16-dimensional space, extracts a high-dimensional representation of the node features, thereby capturing the complex interactions and dependencies between nodes and enhancing the model learning ability; the second convolution layer maps the feature dimension of each node to, for example, a 32-dimensional space, further enriching the feature information of each node; the third convolution layer compresses the feature dimension of the node to, for example, a 16-dimensional space, and is connected to the fully connected layer, reducing the computational complexity of the fully connected layer and better aggregating the features of each node for accurate temperature prediction.

[0052] In some embodiments of the present invention, S201 also includes S10 and S11, which is explained by the following steps.

[0053] S10. Obtain a first adjacency matrix of the directed connection graph and a first degree matrix of each node in the directed connection graph.

[0054] S11. Determine a normalized first Laplacian matrix based on the first adjacency matrix and the first degree matrix.

[0055] In an embodiment of the present invention, for a directed connected graph, the adjacency matrix is ​​a square matrix, the elements of which represent the connection relationship between the vertices in the graph. The degree matrix is ​​used to describe the degree of connection of the nodes in the graph. It is a diagonal matrix, and the elements on its diagonal represent the degree of the corresponding node (that is, the number of edges connected to the node). The first adjacency matrix of the directed connected graph and the first degree matrix of each node are obtained, and the first adjacency matrix, the first degree matrix and the matrix calculation formula are used to calculate to obtain the normalized first Laplace matrix. Among them, the matrix calculation formula (1) is as follows:

[0056] ;

[0057] In the above formula (1), is a normalized Laplace matrix, D is a first degree matrix or a second degree matrix or a third degree matrix, A is a first adjacency matrix or a second adjacency matrix or a third adjacency matrix, I is an identity matrix, and U is an orthogonal matrix composed of the eigenvectors of the normalized Laplace matrix. is the diagonal matrix corresponding to the normalized Laplacian matrix, is the transposed matrix corresponding to the orthogonal matrix.

[0058] It is understandable that in some embodiments of the present invention, since the graph data does not have translation invariance, it is not convenient to define convolution operations on the node domain, so the directed connection graph is converted into a normalized Laplacian matrix through the first adjacency matrix and the first degree matrix. The normalized Laplacian matrix can well maintain the local relationship between nodes, so that the convolution layer can effectively capture and utilize the topological structure information of the graph.

[0059] In some embodiments of the present invention, S201 can be implemented through S2011, which is explained by the following steps.

[0060] S2011. Perform a convolution operation based on the first Laplacian matrix, the directed connection graph, and the first filter to obtain a first feature matrix.

[0061] In some embodiments of the present invention, the feature dimension of each node is mapped to a 16-dimensional space through a Laplacian matrix, a first filter and a convolution formula to obtain a first feature matrix; wherein the convolution formula (2) is:

[0062] In the above formula, y is the first characteristic matrix, X is the nodes contained in the directed connection graph at the current moment, is the filter, are the parameters of the filter.

[0063] (2);

[0064] In the above formula, is the rescaled feature matrix, is the maximum eigenvalue of the normalized Laplacian matrix, T k represents the Chebyshev polynomial, k represents the order of the Chebyshev polynomial, ,in, represents the parameters of the Chebyshev polynomial, is the weight obtained by training the temperature prediction model, and is the vector of Chebyshev coefficients. When K=1, the convolution formula (2) can be converted to formula (3) as follows:

[0065] (3);

[0066] In the above formula, is the normalized Laplacian matrix; for The corresponding degree matrix;

[0067] Since the filter parameters can be shared across the entire graph, the continuous application of this form of filter can effectively convolve the K-order neighborhood of the node. In practical applications, further limiting the number of parameters is beneficial to solving the overfitting problem and minimizing the amount of computation per layer (such as matrix multiplication). Therefore, let , formula (3) is further simplified to formula (4), such as:

[0068] (4);

[0069] In the above formula (4), due to The eigenvalue range is [0, 2]. Repeated application of this operator in graph convolutional neural networks will lead to gradient explosion. To alleviate this problem, the formula is renormalized , combining the simplified first-order approximation with the renormalized adjacency matrix, where is the self-connected adjacency matrix of a given directed graph, which is incorporated into the node representation itself when updating the node representation.

[0070] It is understood that in some embodiments of the present invention, a convolution operation is performed based on the first Laplacian matrix, the directed connection graph and the first filter to obtain a first feature matrix. The method extracts a high-dimensional representation of node features by mapping the 1-dimensional features of each node to a 16-dimensional space, thereby capturing the complex interactions and dependencies between nodes and enhancing the model learning ability.

[0071] In some embodiments of the present invention, S20 to S21 are also included before S202, which is explained by the following steps.

[0072] S20, obtaining a second adjacency matrix of the first characteristic matrix and a second degree matrix of each node in the first characteristic matrix.

[0073] S21. Determine a normalized second Laplacian matrix based on the second adjacency matrix and the second degree matrix.

[0074] In some embodiments of the present invention, after the first convolutional layer outputs the first feature matrix, the second adjacency matrix corresponding to the first feature matrix and the second degree matrix of each node are obtained, and the normalized second Laplacian matrix is ​​calculated by combining the second adjacency matrix, the second degree matrix and the above formula (1).

[0075] In some embodiments of the present invention, S202 can be implemented through S2021, which is explained by the following steps.

[0076] S2021. Perform a convolution operation based on the second Laplacian matrix, the first characteristic matrix, and the second filter to obtain a second characteristic matrix.

[0077] In some embodiments of the present invention, a second characteristic matrix is ​​obtained by performing a convolution operation on the second Laplacian matrix, the first characteristic matrix, the second filter, and the above formulas (2) to (4).

[0078] It is understandable that in some embodiments of the present invention, the second adjacency matrix of the first feature matrix and the second degree matrix of each node are obtained, a normalized second Laplacian matrix is ​​determined based on the second adjacency matrix and the second degree matrix, and a convolution operation is performed based on the second Laplacian matrix, the first feature matrix and the second filter to obtain the second feature matrix. This method not only maintains the local relationship between nodes well through the normalized Laplacian matrix, so that the convolution layer can effectively capture and utilize the topological structure information of the graph, but also further maps the feature dimension of each node from, for example, 16 to, for example, 32-dimensional space, further enriching the feature information of each node.

[0079] In some embodiments of the present invention, S203 also includes S30 to S31, which is explained by the following steps.

[0080] S30, obtaining a third adjacency matrix of the second characteristic matrix and a third degree matrix of each node in the second characteristic matrix.

[0081] S31. Determine a normalized third Laplacian matrix based on a third adjacency matrix and a third degree matrix.

[0082] In some embodiments of the present invention, after the second convolutional layer outputs the second feature matrix, the third adjacency matrix of the second feature matrix and the third degree matrix of each node are obtained, and the normalized third Laplacian matrix is ​​calculated by the third adjacency matrix, the third degree matrix and formula (1).

[0083] In some embodiments of the present invention, S203 may be implemented through S2031, which is explained by the following steps.

[0084] S2031. Perform a convolution operation based on the third Laplacian matrix, the second feature matrix, and the third filter to obtain a target feature matrix.

[0085] In some embodiments of the present invention, a target feature matrix is ​​obtained by performing a convolution operation on the third Laplacian matrix, the second feature matrix, the third filter, and the above formulas (2) to (4).

[0086] It is understandable that in some embodiments of the present invention, the third adjacency matrix of the second feature matrix and the third degree matrix of each node are obtained, a normalized third Laplace matrix is ​​determined based on the third adjacency matrix and the third degree matrix, and a convolution operation is performed based on the third Laplace matrix, the second feature matrix and the third filter to obtain a target feature matrix. This method not only maintains the local relationship between nodes well through the normalized Laplace matrix, so that the convolution layer can effectively capture and utilize the topological structure information of the graph, but also compresses the feature dimension such as 32 to a 16-dimensional space, and connects it to the fully connected layer, reducing the computational complexity of the fully connected layer and better aggregating the features of each node for temperature prediction.

[0087] Reference Figure 4 , shows a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. The specific embodiment of the present invention does not limit the specific implementation of the electronic device.

[0088] like Figure 4 As shown, the electronic device may include: a processor (processor) 502, a communication interface (Communications Interface 504, a memory (memory) 506, and a communication bus 508.

[0089] in:

[0090] The processor 502 , the communication interface 504 , and the memory 506 communicate with each other via a communication bus 508 .

[0091] The communication interface 504 is used to communicate with other electronic devices or servers.

[0092] The processor 502 is used to execute the program 510, and specifically can execute the relevant steps in the above method embodiment.

[0093] Specifically, the program 510 may include program codes, which include computer operation instructions.

[0094] The processor 502 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0095] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0096] The program 510 may be specifically used to enable the processor 502 to execute operations corresponding to the methods described in the above method embodiments.

[0097] The specific implementation of each step in program 510 can refer to the corresponding description of the corresponding steps and units in the above method embodiment, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described devices and modules can refer to the corresponding process description in the above method embodiment, which will not be repeated here.

[0098] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present invention can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present invention.

[0099] The method according to the embodiment of the present invention described above may be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or as computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and will be stored in a local recording medium, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown herein.

[0100] Those of ordinary skill in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present invention.

[0101] The above implementation methods are only used to illustrate the embodiments of the present invention, and are not limitations of the embodiments of the present invention. Ordinary technicians in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the embodiments of the present invention. The patent protection scope of the embodiments of the present invention should be defined by the claims.

Claims

1. A method for predicting coal spontaneous combustion temperature based on graph convolutional neural network, characterized in that: include: Obtain information on various gases produced when coal spontaneously combusts in the air; The temperature prediction model is used to use the multiple gas information to perform temperature prediction to obtain the current coal self-ignition temperature; the temperature prediction model includes an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer and a fully connected layer connected in sequence; the input layer is used to embed the multiple gas information into a directed connection graph; the temperature prediction model is a graph convolutional neural network; each node of the directed connection graph represents each gas in the multiple gas information and an attribute value associated with each gas, and the attribute values ​​include gas concentration, gas oxygen consumption rate, gas generation rate, gas generation rate / gas consumption rate and gas ratio; the edges of the directed connection graph are directed edges, and the directed edges represent the generation relationship between nodes; the gas consumption rate includes the gas oxygen consumption rate and the gas generation rate; the first convolutional layer includes a first filter, the second convolutional layer includes a second filter and the third convolutional layer includes a third filter; the number of the first filters is equal to the number of the third filters, and the number of the second filters is greater than the number of the first filters; The directed connection graph is convolved using the first filter to obtain a first feature matrix, including: Obtaining a first adjacency matrix of the directed connection graph and a first degree matrix of each node in the directed connection graph; the first adjacency matrix is ​​a square matrix, the elements of which represent the connection relationship between the nodes in the directed connection graph; the first degree matrix is ​​used to describe the degree of connection of the nodes in the directed connection graph, the first degree matrix is ​​a diagonal matrix, and the elements on the diagonal line of the diagonal matrix represent the degree of the corresponding node; determining a normalized first Laplacian matrix based on the first adjacency matrix and the first degree matrix; The step of performing a convolution operation on the directed connection graph using the first filter to obtain a first feature matrix specifically includes: A convolution operation is performed based on the first Laplacian matrix, the directed connection graph, and the first filter to obtain the first feature matrix, wherein the first feature matrix is ​​shown in the following formula: ； In the above formula, y is the first characteristic matrix, , D is the first degree matrix, A is the first adjacency matrix, refers to the leading coefficient of the Chebyshev polynomial, refers to the second coefficient of the Chebyshev polynomial, I is the identity matrix; X is a node of the directed connection graph.

2. The method according to claim 1, characterized in that: The step of performing temperature prediction using the multiple gas information through a temperature prediction model to obtain the current coal auto-ignition temperature includes: Inputting the plurality of gas information into the input layer to obtain the directed connection graph; Based on the first filter, the second filter and the third filter, a convolution operation is performed on the directed connection graph in sequence to obtain a target feature matrix; The target feature matrix is ​​fitted through the fully connected layer to obtain the auto-ignition temperature.

3. The method according to claim 2, characterized in that The step of sequentially performing convolution operations on the directed connection graph based on the first filter, the second filter, and the third filter to obtain a target feature matrix includes: Using the first filter to perform a convolution operation on the directed connection graph to obtain a first feature matrix; Using the second filter to perform a convolution operation on the first feature matrix to obtain a second feature matrix; The third filter is used to perform a convolution operation on the second feature matrix to obtain the target feature matrix.

4. The method according to claim 3, characterized in that Before the convolution operation is performed on the first feature matrix by using the second filter to obtain the second feature matrix, the method further includes: Obtaining a second adjacency matrix of the first characteristic matrix and a second degree matrix of each node in the first characteristic matrix; determining a normalized second Laplacian matrix based on the second adjacency matrix and the second degree matrix; The using the second filter to perform a convolution operation on the first feature matrix to obtain a second feature matrix specifically includes: A convolution operation is performed based on the second Laplacian matrix, the first characteristic matrix, and the second filter to obtain the second characteristic matrix.

5. The method according to claim 3, characterized in that: Before the convolution operation is performed on the second feature matrix by using the third filter to obtain the target feature matrix, the method further includes: Obtaining a third adjacency matrix of the second characteristic matrix and a third degree matrix of each node in the second characteristic matrix; determining a normalized third Laplacian matrix based on the third adjacency matrix and the third degree matrix; The using the third filter to perform a convolution operation on the second feature matrix to obtain the target feature matrix specifically includes: A convolution operation is performed based on the third Laplacian matrix, the second feature matrix, and the third filter to obtain the target feature matrix.

Images