Mine microseismic early warning method, device and equipment based on spatiotemporal graph neural network

Through the method based on the spatio-temporal graph neural network, a mine microseismic early warning system was constructed, which solved the problems of insufficient microseismic data utilization and poor model adaptability in traditional methods, achieved higher prediction accuracy and timeliness, and provided more reliable mine safety Assure.

CN118915143BActive Publication Date: 2024-12-06CENT SOUTH UNIV
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411406042.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-10
Publication Date
2024-12-06
Estimated Expiration
2044-10-10

AI Technical Summary

Technical Problem

Traditional empirical-driven methods and data-driven machine learning methods are difficult to effectively utilize microseismic data, and cannot accurately predict microseismic events in mines. The dynamic update capability of the model is weak, making it difficult to adapt to changes in mine geological conditions and mining activities.

Method used

Using a method based on the spatiotemporal graph neural network, by obtaining microseismic information data, dividing grid nodes, building a spatiotemporal graph neural network, combining attention mechanism and LSTM model, the impact factors of microseismic events are updated in real time, and early warning output is generated to improve the accuracy of prediction. and timeliness.

Benefits of technology

It significantly improves the accuracy and timeliness of mine microseismic warnings, enhances the adaptability and dynamic update capabilities of the model, can better capture the correlation between microseismic events and geological structure, and provide more reliable security guarantees.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915143B_ABST
    Figure CN118915143B_ABST
Patent Text Reader

Abstract

The present invention provides a method, device and equipment for early warning of microseismic events in mines based on spatiotemporal graph neural networks, and relates to the technical field of graph neural networks. By describing each area of ​​the mine as a node network, linking each other to form an associated graph, the correlation between each node based on the impact of microseismic events is explored. Based on microseismic events occurring at different times, a three-factor weight based on time, space and strong information is formed, and after the attention weight of GAT is updated, a multi-head attention mechanism is used to obtain the weight of the edge between the final nodes. After being processed by the spatiotemporal graph neural network, the number of microseismic events, average energy and maximum energy occurring in each node in the next time period are predicted, and an early warning is given for possible safety accidents in mining. The correlation between microseismic events and geological structures is captured in the above manner, and the weights are dynamically updated in real time, with good generalization performance and dynamic update capabilities, thereby improving the accuracy and timeliness of the prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of graph neural network, and in particular to a method, device and equipment for early warning of microseismic events in mines based on spatiotemporal graph neural network. Background Art

[0002] The traditional experience-driven method does not make full use of large-scale microseismic data, cannot reliably find the actual laws of microseismic events, and is difficult to achieve rapid and effective early warning of mine earthquakes. The geological conditions of mines are complex and changeable. Microseismic activity is not only related to geological structures, but also the dynamic impact of mining activities on rock stress and the propagation and mutual influence of microseismic events between rock masses, which makes the influencing factors of microseismic events in the same area also have a dynamic change process. Most of the existing data-driven machine learning methods only analyze the clustering of microseismic events, lack a description from a holistic perspective, have poor generalization performance, and weak dynamic model update capabilities. Therefore, it is difficult to adapt to sudden changes in mine geological conditions or adjustments in mining activities, and less consideration is given to the mutual propagation and influence between microseismic events, resulting in low prediction accuracy and short timeliness.

[0003] Therefore, how to improve the adaptability and generalization ability of machine learning models to cope with the complex and changeable situations of mine geological conditions and mining activities, thereby improving the accuracy and timeliness of predictions has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In order to significantly improve the accuracy and timeliness of early warning and provide more reliable protection for mine safety, the present application provides a mine microseismic early warning method, device and equipment based on spatiotemporal graph neural network.

[0005] In the first aspect, the present application provides a method for early warning of microseismic events in mines based on a spatiotemporal graph neural network, which adopts the following technical solutions:

[0006] A mine microseismic early warning method based on spatiotemporal graph neural network, comprising:

[0007] Acquire the mine microseismic information data of the mining area to be predicted from the microseismic monitoring system and its own data processing software, wherein the mine microseismic information data at least includes the information of microseismic events, spatial information data and microseismic energy, and the information of microseismic events includes the time of microseismic events and the number of microseismic events;

[0008] The mining area to be predicted is divided into several grids of the same shape according to the geological characteristics of the ore body and the geological structure of the mining area. The divided grids cover the entire mining area to be predicted, and the center of each grid constitutes a node of a graph, and different nodes are connected to form edges;

[0009] The information of microseismic events belonging to each node grid is processed in an isochronous manner, and the number of microseismic events, average microseismic energy, maximum microseismic energy and minimum microseismic energy per hour of each grid are summarized and normalized;

[0010] A spatiotemporal graph neural network is constructed based on the spatial information data corresponding to each node, the processed microseismic event information, and the processed microseismic energy;

[0011] Construct an attention mechanism based on real-time updates of microseismic events and combine it with the spatiotemporal graph neural network to generate GAT output;

[0012] Processing the GAT output through an LSTM model to obtain an LSTM output result;

[0013] The results output by the LSTM are denormalized to obtain the number of microseismic events, average microseismic energy, and maximum microseismic energy of each node in the next time period;

[0014] The risk level is determined based on the number of microseismic events, the average microseismic energy value, and the maximum microseismic energy value in the next time period of each node, so as to provide real-time warning for the mining area to be predicted.

[0015] Optionally, the construction of an attention mechanism based on real-time updates of microseismic events and combining the spatiotemporal graph neural network to generate GAT outputs includes:

[0016] Determine the time impact factor, space impact factor and energy impact factor based on microseismic events;

[0017] Determine the weights of the three factors according to the time impact factor, the space impact factor and the energy impact factor;

[0018] Update the attention coefficient according to the three-factor weights, and normalize the attention coefficient through the softmax function to obtain the attention weight of the spatiotemporal graph neural network;

[0019] Based on the attention weights, edge weights based on a multi-head attention mechanism are obtained to generate GAT outputs.

[0020] Optionally, the determining of the time influence factor, the space influence factor and the energy influence factor based on the microseismic event includes:

[0021] Based on the microseismic information data, the earthquake source is obtained Or the spatial coordinate information x of node a a ,y a ,z a , source The energy size and the source The time of earthquake ;

[0022] The distance from source a to node i is defined as ,in

[0023] ;

[0024] According to the source The distance to node i is used to calculate the travel time from the time of microseismic occurrence to node i. , the calculation formula is:

[0025]

[0026] Where v is the constant wave velocity in the entire area;

[0027] Calculate the current time and earthquake source The time interval between earthquake occurrences is , the calculation formula is:

[0028]

[0029] Among them, t is the current time point;

[0030] Determine the time impact factor of microseismic event a on node i as follows: ,

[0031]

[0032] in, For the earthquake source After the seismic energy is transmitted to node i, the time loss coefficient at the time point is: After the energy propagates to the node, the energy in the node area per second in the metal ore medium affects the attenuation coefficient Determined, , attenuation coefficient It can be calculated or directly obtained from the data processing software that comes with the microseismic monitoring system.

[0033] Determine the energy impact factor of microseismic event a on node i as follows: ,

[0034]

[0035] in, is the energy loss coefficient of the seismic energy transmitted from source a to node i, , the loss rate caused by the energy conduction per meter in the ore body medium of the metal mine Determined, , loss rate It can be calculated or directly obtained from the data processing software that comes with the microseismic monitoring system;

[0036] Define the spatial attribution factor of node j to node i based on microseismic event a as ,

[0037]

[0038] in, and are the distances from a to nodes i and j respectively.

[0039] The spatial distance factor of node j to node i based on microseismic event a is defined as ,

[0040]

[0041] in, is the mean distance from node i to all nodes, is the distance between node i and node j;

[0042] The spatial impact factor of microseismic event a on node j is defined as , the calculation formula is as follows:

[0043] ;

[0044] Determine the spatial impact factor of node j on node i based on microseismic event a: , the calculation formula is as follows:

[0045] .

[0046] Optionally, the three-factor weights The calculation formula is:

[0047] ;

[0048] exist When, yes Normalize and get the three-factor weight of node j to node i , the calculation formula is:

[0049] .

[0050] Optionally, the updated attention coefficient The formula is:

[0051] ,

[0052] in, It is a nonlinear activation function used to improve the expressiveness of the model. is the set of all microseismic events in the time period, and the impact of all microseismic events is updated through the summation function. is a dimensional learnable parameter vector, and is the feature vector of nodes i and j in the input layer, with an initial dimension of F, is a learnable weight matrix that shares a linear transformation, , are the input and output dimensions, respectively, Represents the vector concatenation operation, which concatenates the vectors into Dimension, then dot multiplication , mapping high-dimensional features to a real number range;

[0053] The formula for normalizing the attention coefficient through the softmax function is:

[0054]

[0055] in, is the average distance from i to all nodes, , is the set of neighbor nodes linked to node i.

[0056] Optionally, the GAT output is processed by an LSTM model to obtain a result of the LSTM output, including:

[0057] The GAT output is passed into LSTM as input, and the data is calculated through the forget gate, input gate, and output gate to obtain the LSTM output result.

[0058] Optionally, the forget gate is ,in, is the output of the forget gate; It is a sigmoid function, which is used to output a value between 0 and 1, indicating the degree of information retention; is the weight matrix, used to adjust the importance of the input; is the output of the previous time step, The input of the current time step is used to concatenate into a new vector so that the forget gate can integrate the past state information and the current input at the same time; is a bias term used to adjust the output range of the sigmoid function;

[0059] The input gate is ,in, , , and are the input gate and candidate memory cell states respectively; tanh is the activation function used to map the input to the range of -1 to 1; and is the weight matrix; and is the bias term; and are the memory cell states of the current time step and the previous time step respectively;

[0060] The output gate is ,in, , is the output of the output gate; is the hidden state at the current time step; is the weight matrix; is the bias term.

[0061] In a second aspect, the present application provides a mine microseismic early warning device based on a spatiotemporal graph neural network, and the mine microseismic early warning device based on a spatiotemporal graph neural network includes:

[0062] A data acquisition module is used to obtain the mine microseismic information data of the mining area to be predicted from the microseismic monitoring system and its own data processing software, wherein the mine microseismic information data at least includes the information of microseismic events, spatial information data and microseismic energy, and the information of microseismic events includes the time of microseismic events and the number of microseismic events;

[0063] A grid division module is used to divide the mining area to be predicted into several grids of the same shape according to the geological characteristics of the ore body and the geological structure of the mining area, wherein the divided grids cover the entire mining area to be predicted, each grid center constitutes a node of a graph, and different nodes are connected to form edges;

[0064] The data processing module is used to perform isochronous processing on the information of microseismic events belonging to each node grid, and summarize the number of microseismic events, average microseismic energy, maximum microseismic energy and minimum microseismic energy of each grid per hour for normalization processing;

[0065] A spatiotemporal graph neural network construction module is used to construct a spatiotemporal graph neural network based on the spatial information data corresponding to each node, the processed microseismic event information, and the processed microseismic energy;

[0066] An output module is used to construct an attention mechanism based on real-time updates of microseismic events and generate GAT output in combination with the spatiotemporal graph neural network;

[0067] An output processing module, used for processing the GAT output through an LSTM model to obtain an LSTM output result;

[0068] A denormalization module is used to denormalize the results output by the LSTM to obtain the number of microseismic events, average microseismic energy and maximum microseismic energy of each node in the next time period;

[0069] The microseismic early warning module is used to judge the risk level according to the number of microseismic events, average microseismic energy value and maximum microseismic energy value in the next time period of each node, so as to provide real-time early warning for the mining area to be predicted.

[0070] Optionally, the output module is used to determine a time impact factor, a space impact factor and an energy impact factor based on the microseismic event;

[0071] Determine the weights of the three factors according to the time impact factor, the space impact factor and the energy impact factor;

[0072] Update the attention coefficient according to the three-factor weights, and normalize the attention coefficient through the softmax function to obtain the attention weight of the spatiotemporal graph neural network;

[0073] Based on the attention weights, edge weights based on a multi-head attention mechanism are obtained to generate GAT outputs.

[0074] In a third aspect, the present application provides a mine microseismic early warning device based on a spatiotemporal graph neural network, the device comprising: a memory and a processor, the processor executing the method described above when running the mine microseismic early warning program based on a spatiotemporal graph neural network stored in the memory.

[0075] In summary, the present application includes the following beneficial technical effects:

[0076] This application describes each area of ​​the mine as a node network, which is linked to each other to form an associated graph, and explores the correlation between each node based on the impact of microseismic events. Based on microseismic events occurring at different times, a three-factor weight based on time, space, and strong information is formed, and the attention weight is updated through GAT. The multi-head attention mechanism is used to obtain the weight of the edge between the final nodes. After processing by the spatiotemporal graph neural network, the number of microseisms, average energy, and maximum energy occurring in each node in the next time period are predicted, and possible safety accidents in mining are warned. The correlation between microseismic events and geological structures is captured in the above manner, and the weights are updated dynamically in real time, with good generalization performance and dynamic update capabilities, thereby improving the accuracy and timeliness of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is a schematic diagram of the structure of a mine microseismic early warning device based on a spatiotemporal graph neural network in the hardware operating environment involved in the embodiment of the present application;

[0078] Figure 2 It is a flow chart of the first embodiment of the mine microseismic early warning method based on the spatiotemporal graph neural network of the present application;

[0079] Figure 3 It is a schematic diagram of the overall process of the early warning method in the mine microseismic early warning method based on the spatiotemporal graph neural network of this application;

[0080] Figure 4 It is a structural block diagram of the first embodiment of the mine microseismic early warning device based on the spatiotemporal graph neural network of the present application. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below through the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0082] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of a mine microseismic early warning device based on a spatiotemporal graph neural network in the hardware operating environment involved in the embodiment of the present application.

[0083] like Figure 1 As shown, the mine microseismic early warning device based on the spatiotemporal graph neural network may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0084] Those skilled in the art will understand that Figure 1 The structure shown in does not constitute a limitation on the mine microseismic early warning device based on spatiotemporal graph neural network, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0085] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a mine microseismic early warning program based on a spatiotemporal graph neural network.

[0086] exist Figure 1 In the mine microseismic early warning device based on the space-time graph neural network shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the present application can be set in the mine microseismic early warning device based on the space-time graph neural network, and the mine microseismic early warning device based on the space-time graph neural network calls the mine microseismic early warning program based on the space-time graph neural network stored in the memory 1005 through the processor 1001, and executes the mine microseismic early warning method based on the space-time graph neural network provided in the embodiment of the present application.

[0087] The present application embodiment provides a mine microseismic early warning method based on a spatiotemporal graph neural network, referring to Figure 2 , Figure 2 This is a flow chart of the first embodiment of the mine microseismic early warning method based on the spatiotemporal graph neural network of the present application.

[0088] This application is based on the actual scenario of mining. Taking into account the non-independence of microseismic events in mines, the microseismic energy of a single area is affected by the energy propagation of each area of ​​the mine. It is a systematic and holistic problem. Therefore, each area of ​​the mine is described as a node network, which is linked to each other to form a related graph to explore the correlation between the nodes based on the impact of microseismic events. By discussing the microseismic events occurring at different times, a three-factor weight based on time, space, and strength (energy intensity) information is formed, and the attention weights of GAT are updated. The multi-head attention mechanism is used to obtain the weights of the edges between the final nodes. After being processed by the spatiotemporal graph neural network, the number of microseisms, average energy, and maximum energy occurring in each node in the next time period are predicted, and warnings are issued for possible safety accidents in mining. Since this method fully considers the impact of the continuous occurrence of microseismic events, captures the correlation between microseismic events and geological structures, and dynamically updates weights in real time, it has good generalization performance and dynamic update capabilities. The overall implementation process of the early warning method can refer to Figure 3 shown.

[0089] In this embodiment, the mine microseismic early warning method based on the spatiotemporal graph neural network includes the following steps:

[0090] Step S10: Acquire the mine microseismic information data of the mining area to be predicted from the microseismic monitoring system and its own data processing software.

[0091] It should be noted that the mine microseismic information data at least includes information on microseismic events, spatial information data and microseismic energy. The information on microseismic events includes the time of microseismic events and the number of microseismic events. The mine microseismic information data can be obtained from the microseismic monitoring system and its own data processing software.

[0092] Step S20: Divide the mining area to be predicted into a number of grids of the same shape according to the geological characteristics of the ore body and the geological structure of the mining area.

[0093] In specific implementation, the mining area to be predicted can be divided into grids according to the geological characteristics of the ore body and the geological structure of the mining area. For example, it can be divided into n grids of the same shape. The number of grids can be set according to demand, wherein the divided grids cover the entire mining area to be predicted, and each grid center constitutes a node of a graph, and different nodes are connected into edges.

[0094] Step S30: isochronize the information of microseismic events belonging to each node grid, and summarize the number of microseismic events, average microseismic energy, maximum microseismic energy and minimum microseismic energy of each grid per hour for normalization.

[0095] It should be noted that the microseismic information data of the mine in each node grid can be obtained by using the grid divided as described above, and then the microseismic information data of the mine belonging to each node grid is processed accordingly. Specifically, the information of the microseismic events is processed in an isochronous manner, and the average microseismic energy, maximum microseismic energy and minimum microseismic energy can be obtained through the above-mentioned microseismic energy. Then, the number of microseismic events per hour, the average microseismic energy, the maximum microseismic energy and the minimum microseismic energy of each grid are summarized and normalized.

[0096] Step S40: construct a spatiotemporal graph neural network according to the spatial information data corresponding to each node, the processed microseismic event information, and the processed microseismic energy.

[0097] After the above data is processed, it is combined with the spatial information data corresponding to each node and input into the preset spatio-temporal graph neural network to complete the construction. The spatio-temporal graph neural network (ST-GNNs) used in this embodiment is a deep learning model.

[0098] Step S50: construct an attention mechanism based on real-time updates of microseismic events, and generate GAT output in combination with the spatiotemporal graph neural network.

[0099] It should be noted that constructing an attention mechanism based on real-time updates of microseismic events and generating GAT output in combination with the spatiotemporal graph neural network includes: determining the time influence factor, space influence factor and energy influence factor based on microseismic events; determining the weights of the three factors according to the time influence factor, the space influence factor and the energy influence factor; updating the attention coefficient according to the three-factor weights, and normalizing the attention coefficient through the softmax function to obtain the attention weight of the spatiotemporal graph neural network; obtaining the edge weight based on the multi-head attention mechanism based on the attention weight to generate GAT output.

[0100] It should be noted that the latest microseismic event source information obtained ( , , , , ), find the node Based on the microseismic event a Impact Factor .in, , , For the earthquake source or the spatial coordinate information of node a, For the earthquake source The energy size, For the earthquake source Time of earthquake.

[0101] Further, based on the microseismic information data, the earthquake source is obtained. Or the spatial coordinate information of node a , , , source The energy size and the source The time of earthquake ;

[0102] The distance from source a to node i is defined as in,

[0103] ;

[0104] According to the source The distance to node i is used to calculate the travel time from the time of microseismic occurrence to node i. , the calculation formula is:

[0105]

[0106] Where v is the constant wave velocity in the entire area; calculate the current time and the earthquake source The time interval between earthquake occurrences is The calculation formula is:

[0107]

[0108] Where t is the current time point; the time impact factor of microseismic event a on node i is determined as ,in,

[0109]

[0110] in, For the earthquake source After the seismic energy is transmitted to node i, the time loss coefficient at the time point is: After the energy propagates to the node, the energy in the node area per second in the metal ore medium affects the attenuation coefficient Determined, .

[0111] Determine the energy impact factor of microseismic event a on node i as follows: ,

[0112]

[0113] in, is the energy loss coefficient of the seismic energy transmitted from source a to node i, , the loss rate caused by the energy conduction per meter in the ore body medium of the metal mine Determined, The spatial attribution factor of node j to node i based on microseismic event a is defined as ,in,

[0114]

[0115] in, and are the distances from a to nodes i and j respectively. It should be noted that when a is closer to j, >1, the source a is more likely to belong to node j. The closer the source a is to node j, the greater the influence of node j on node i through a. Since L is limited to a real number not less than 1, it effectively prevents Unlimited increase.

[0116] The spatial distance factor of node j to node i based on microseismic event a is defined as ,

[0117]

[0118] in, is the mean distance from node i to all nodes, is the distance between node i and node j. It should be noted that the fraction represents that the shorter the distance between nodes i and j, the greater the mutual influence. Since this effect is more important and in order to make a clear distinction between short distance and long distance, the square term is used to process the fraction.

[0119] The spatial impact factor of microseismic event a on node j is defined as , the calculation formula is as follows:

[0120] ;

[0121] Determine the spatial impact factor of node j on node i based on microseismic event a: , when the earthquake source a is far away from node j, it obviously has little effect on node j, and the coefficient It may be less than 1. When the distance is closer, a larger spatial influence coefficient will appear. Limiting the distance L to a real number not less than 1 also prevents the coefficient from increasing infinitely. The calculation formula is as follows:

[0122] .

[0123] Furthermore, the three-factor weights The calculation formula is:

[0124] .

[0125] It should be noted that the impact of node j on node i based on microseismic event a is affected by multiple factors such as source time, space, and energy. The greater the energy, the greater the impact; the closer the source a is to node j, the greater the impact; the closer the distance between node i and node j, the greater the impact; after the energy of source a is transmitted to node i, the impact will continue to decrease with the increase of time; the longer the energy propagation path of source a, the more loss it suffers, and the smaller the impact.

[0126] exist When, yes Normalize and get the three-factor weight of node j to node i , the calculation formula is: .

[0127] After obtaining the above three-factor weights, the attention coefficient can be updated as follows: ;

[0128] in, It is a nonlinear activation function used to improve the expressiveness of the model. is the set of all microseismic events in the time period, and the impact of all microseismic events is updated through the summation function. is a dimensional learnable parameter vector, and is the feature vector of nodes i and j in the input layer, with an initial dimension of F, is a learnable weight matrix that shares a linear transformation, , are the input and output dimensions, respectively, Represents the vector concatenation operation, which concatenates the vectors into Dimension, then dot multiplication , mapping high-dimensional features to a real number range.

[0129] Furthermore, in order to facilitate the comparison of attention coefficients of different nodes, the attention coefficients are normalized by the softmax function to obtain the attention weights.

[0130] in, is the average distance from i to all nodes, , is the set of neighboring nodes linked to node i. It should be noted that in this embodiment, a distance factor is added to the above formula. , which is used to enhance the expression of mutual influence relationship between nodes with similar distances. The closer the distance between nodes i and j, the greater the influence.

[0131] It should be noted that the above attention weights are also the edge weights of the multi-head attention mechanism, based on which the GAT output can be generated. It should be emphasized that the feature vector output after information aggregation can be obtained from the attention weights .

[0132]

[0133] in, is the activation function, and the formula indicates that the output information of node i is aggregated by the mutual influence of multiple nodes.

[0134] In order to prevent overfitting, a multi-head attention mechanism is used to introduce multiple groups Vector Sum Matrix to make the model more stable and get the final output .

[0135]

[0136] in, is the vector of the node after the attention weight and multi-head attention mechanism are updated, which aggregates the information of each node and the microseismic event for node i. K is the number of attention heads, which represents the use of K groups Vector Sum Matrix to get the new vector of node i .

[0137] Step S60: Process the GAT output through an LSTM model to obtain an LSTM output result.

[0138] It should be noted that the output of the GAT network that updates the attention weight based on microseismic events is processed by the LSTM model to obtain the final output result. Specifically, the GAT output is passed into the LSTM as input, and the data is calculated through the forget gate, input gate, and output gate to obtain the LSTM output result.

[0139] The forget gate is ,in, is the output of the forget gate; It is a sigmoid function, which is used to output a value between 0 and 1, indicating the degree of information retention; is the weight matrix, used to adjust the importance of the input; is the output of the previous time step, The input of the current time step is used to concatenate into a new vector so that the forget gate can integrate the past state information and the current input at the same time; It is a bias term used to adjust the output range of the sigmoid function. It should be noted that the forget gate determines the amount of information discarded from the memory unit.

[0140] The input gate is ,in, , and are the input gate and candidate memory cell states respectively; tanh is the activation function used to map the input to the range of -1 to 1; and is the weight matrix; and is the bias term; and The input gate determines what new information will be added to the memory cell.

[0141] The output gate is ,in, , is the output of the output gate; is the hidden state at the current time step; is the weight matrix; The output gate determines the amount of information output from the memory cell.

[0142] Step S70: Denormalize the result output by the LSTM to obtain the number of microseismic events, average microseismic energy and maximum microseismic energy of each node in the next time period.

[0143] Step S80: Determine the risk level according to the number of microseismic events, the average microseismic energy value, and the maximum microseismic energy value in the next time period of each node, so as to provide real-time warning for the mining area to be predicted.

[0144] This embodiment describes each area of ​​the mine as a node network, which is linked to each other to form an associated graph, and explores the correlation between each node based on the impact of microseismic events. Based on the microseismic events occurring at different times, a three-factor weight based on time, space, and strong information is formed, and the attention weight is updated through GAT. The multi-head attention mechanism is used to obtain the weight of the edge between the final nodes. After being processed by the spatiotemporal graph neural network, the number of microseisms, average energy, and maximum energy occurring in each node in the next time period are predicted, and an early warning is issued for possible safety accidents in mining. The correlation between microseismic events and geological structures is captured in the above manner, and the weights are dynamically updated in real time, which has good generalization performance and dynamic update capabilities, thereby improving the accuracy and timeliness of the prediction.

[0145] Reference Figure 4 , Figure 4 This is a structural block diagram of the first embodiment of the mine microseismic early warning device based on the spatiotemporal graph neural network of the present application.

[0146] like Figure 4 As shown, the mine microseismic early warning device based on the spatiotemporal graph neural network proposed in the embodiment of the present application includes:

[0147] The data acquisition module 10 is used to obtain the mine microseismic information data of the mining area to be predicted from the microseismic monitoring system and its own data processing software. The mine microseismic information data at least includes the information of microseismic events, spatial information data and microseismic energy. The information of microseismic events includes the time of microseismic events and the number of microseismic events.

[0148] The grid division module 20 is used to divide the mining area to be predicted into a number of grids of the same shape according to the geological characteristics of the ore body and the geological structure of the mining area, wherein the divided grids cover the entire mining area to be predicted, each grid center constitutes a node of a graph, and different nodes are connected to form edges.

[0149] The data processing module 30 is used to perform isochronous processing on the information of microseismic events belonging to each node grid, and summarize the number of microseismic events, average microseismic energy, maximum microseismic energy and minimum microseismic energy of each grid per hour for normalization processing.

[0150] The spatiotemporal graph neural network construction module 40 is used to construct a spatiotemporal graph neural network according to the spatial information data corresponding to each node, the processed microseismic event information and the processed microseismic energy.

[0151] The output module 50 is used to construct an attention mechanism based on real-time updates of microseismic events and generate GAT output in combination with the spatiotemporal graph neural network.

[0152] The output processing module 60 is used to process the GAT output through the LSTM model to obtain the LSTM output result.

[0153] The denormalization module 70 is used to denormalize the results output by the LSTM to obtain the number of microseismic events, average microseismic energy and maximum microseismic energy of each node in the next time period.

[0154] The microseismic early warning module 80 is used to determine the risk level according to the number of microseismic events, the average microseismic energy value, and the maximum microseismic energy value in the next time period of each node, so as to provide real-time early warning for the mining area to be predicted.

[0155] It should be understood that the above is only an example and does not constitute any limitation on the technical solution of the present application. In specific applications, technicians in this field can make settings as needed, and the present application does not impose any limitation on this.

[0156] This embodiment describes each area of ​​the mine as a node network, which is linked to each other to form an associated graph, and explores the correlation between each node based on the impact of microseismic events. Based on the microseismic events occurring at different times, a three-factor weight based on time, space, and strong information is formed, and the attention weight is updated through GAT. The multi-head attention mechanism is used to obtain the weight of the edge between the final nodes. After being processed by the spatiotemporal graph neural network, the number of microseisms, average energy, and maximum energy occurring in each node in the next time period are predicted, and an early warning is issued for possible safety accidents in mining. The correlation between microseismic events and geological structures is captured in the above manner, and the weights are dynamically updated in real time, which has good generalization performance and dynamic update capabilities, thereby improving the accuracy and timeliness of the prediction.

[0157] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present application. In practical applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0158] In addition, for technical details not described in detail in this embodiment, please refer to the method for mine microseismic early warning based on spatiotemporal graph neural network provided in any embodiment of the present application, which will not be repeated here.

[0159] In addition, it should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0160] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0161] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory (ROM) / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0162] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A mine microseismic early warning method based on spatiotemporal graph neural network, characterized in that: include: Acquire the mine microseismic information data of the mining area to be predicted from the microseismic monitoring system and its own data processing software, wherein the mine microseismic information data at least includes the information of microseismic events, spatial information data and microseismic energy, and the information of microseismic events includes the time of microseismic events and the number of microseismic events; The mining area to be predicted is divided into several grids of the same shape according to the geological characteristics of the ore body and the geological structure of the mining area. The divided grids cover the entire mining area to be predicted, and the center of each grid constitutes a node of a graph, and different nodes are connected to form edges; The information of microseismic events belonging to each node grid is processed in an isochronous manner, and the number of microseismic events, average microseismic energy, maximum microseismic energy and minimum microseismic energy per hour of each grid are summarized and normalized; A spatiotemporal graph neural network is constructed based on the spatial information data corresponding to each node, the processed microseismic event information, and the processed microseismic energy; Construct an attention mechanism based on real-time updates of microseismic events and combine it with the spatiotemporal graph neural network to generate GAT output; Processing the GAT output through an LSTM model to obtain an LSTM output result; The results output by the LSTM are denormalized to obtain the number of microseismic events, average microseismic energy, and maximum microseismic energy of each node in the next time period; The risk level is determined based on the number of microseismic events, the average microseismic energy value, and the maximum microseismic energy value in the next time period of each node, so as to provide real-time warning for the mining area to be predicted.

2. The mine microseismic early warning method based on spatiotemporal graph neural network according to claim 1 is characterized in that: The construction of an attention mechanism based on real-time updates of microseismic events and combining the spatiotemporal graph neural network to generate GAT outputs includes: Determine the time impact factor, space impact factor and energy impact factor based on microseismic events; Determine the weights of the three factors according to the time impact factor, the space impact factor and the energy impact factor; Update the attention coefficient according to the three-factor weights, and normalize the attention coefficient through the softmax function to obtain the attention weight of the spatiotemporal graph neural network; Based on the attention weights, edge weights based on a multi-head attention mechanism are obtained to generate GAT outputs.

3. The mine microseismic early warning method based on spatiotemporal graph neural network according to claim 2 is characterized in that: The determination of the time influence factor, the space influence factor and the energy influence factor based on the microseismic event includes: Based on the microseismic information data, the earthquake source is obtained Or the spatial coordinate information of node a , , , source The energy size and the source The time of earthquake ; The distance from source a to node i is defined as ,in ; According to the source The distance to node i is used to calculate the travel time from the time of microseismic occurrence to node i. , the calculation formula is: Where v is the constant wave velocity in the entire area; Calculate the current time and earthquake source The time interval between earthquake occurrences is , the calculation formula is: Among them, t is the current time point; Determine the time impact factor of microseismic event a on node i as follows: , in, For the source After the seismic energy is transmitted to node i, the time loss coefficient at the time point is: After the energy propagates to the node, the energy in the node area per second in the metal ore medium affects the attenuation coefficient Determined, ; Determine the energy impact factor of microseismic event a on node i as follows: , in, is the energy loss coefficient of the seismic energy transmitted from source a to node i, , the loss rate caused by the energy conduction per meter in the ore body medium of the metal mine Determined, ; Define the spatial attribution factor of node j to node i based on microseismic event a as , in, and are the distances from a to nodes i and j respectively; The spatial distance factor of node j to node i based on microseismic event a is defined as , in, is the mean distance from node i to all nodes, is the distance between node i and node j; The spatial impact factor of microseismic event a on node j is defined as , the calculation formula is as follows: ; Determine the spatial impact factor of node j on node i based on microseismic event a: , the calculation formula is as follows: 。 4. The mine microseismic early warning method based on spatiotemporal graph neural network according to claim 3 is characterized in that: The three-factor weights The calculation formula is: ; exist When, yes Normalize and get the three-factor weight of node j to node i , the calculation formula is: 。 5. The mine microseismic early warning method based on spatiotemporal graph neural network according to claim 4 is characterized in that: The updated attention coefficient The formula is: , in, is a nonlinear activation function used to improve the expressiveness of the model. is the set of all microseismic events in the time period, and the impact of all microseismic events is updated through the summation function. is a dimensional learnable parameter vector, and is the feature vector of nodes i and j in the input layer, with an initial dimension of F, is a learnable weight matrix that shares a linear transformation, , are the input and output dimensions, respectively, Represents the vector concatenation operation, which concatenates the vectors into Dimension, then dot multiplication , mapping high-dimensional features to a real number range; The formula for normalizing the attention coefficient through the softmax function is: in, is the average distance from i to all nodes, , is the set of neighbor nodes linked to node i.

6. The mine microseismic early warning method based on spatiotemporal graph neural network according to claim 1 is characterized in that: The GAT output is processed by the LSTM model to obtain the LSTM output result, including: The GAT output is passed into LSTM as input, and the data is calculated through the forget gate, input gate, and output gate to obtain the LSTM output result.

7. The mine microseismic early warning method based on spatiotemporal graph neural network according to claim 6 is characterized in that: The forget gate is ,in, is the output of the forget gate; It is a sigmoid function, which is used to output a value between 0 and 1, indicating the degree of information retention; is the weight matrix, used to adjust the importance of the input; is the output of the previous time step, The input of the current time step is used to concatenate into a new vector so that the forget gate can integrate the past state information and the current input at the same time; is a bias term used to adjust the output range of the sigmoid function; The input gate is ,in, , , and are the input gate and candidate memory cell states respectively; tanh is the activation function used to map the input to the range of -1 to 1; and is the weight matrix; and is the bias term; and are the memory unit states of the current time step and the previous time step respectively; The output gate is ,in, , is the output of the output gate; is the hidden state at the current time step; is the weight matrix; is the bias term.

8. A mine microseismic early warning device based on spatiotemporal graph neural network, characterized in that: The mine microseismic early warning based on spatiotemporal graph neural network includes: A data acquisition module is used to obtain the mine microseismic information data of the mining area to be predicted from the microseismic monitoring system and its own data processing software, wherein the mine microseismic information data at least includes the information of microseismic events, spatial information data and microseismic energy, and the information of microseismic events includes the time of microseismic events and the number of microseismic events; A grid division module is used to divide the mining area to be predicted into several grids of the same shape according to the geological characteristics of the ore body and the geological structure of the mining area, wherein the divided grids cover the entire mining area to be predicted, each grid center constitutes a node of a graph, and different nodes are connected to form edges; The data processing module is used to perform isochronous processing on the information of microseismic events belonging to each node grid, and summarize the number of microseismic events, average microseismic energy, maximum microseismic energy and minimum microseismic energy of each grid per hour for normalization processing; A spatiotemporal graph neural network construction module is used to construct a spatiotemporal graph neural network based on the spatial information data corresponding to each node, the processed microseismic event information, and the processed microseismic energy; An output module is used to construct an attention mechanism based on real-time updates of microseismic events and generate GAT output in combination with the spatiotemporal graph neural network; An output processing module, used for processing the GAT output through an LSTM model to obtain an LSTM output result; A denormalization module is used to denormalize the results output by the LSTM to obtain the number of microseismic events, average microseismic energy and maximum microseismic energy of each node in the next time period; The microseismic early warning module is used to judge the risk level according to the number of microseismic events, average microseismic energy value and maximum microseismic energy value in the next time period of each node, so as to provide real-time early warning for the mining area to be predicted.

9. The mine microseismic early warning device based on spatiotemporal graph neural network according to claim 8 is characterized in that: The output module is used to determine the time impact factor, space impact factor and energy impact factor based on the microseismic event; Determine the weights of the three factors according to the time impact factor, the space impact factor and the energy impact factor; Update the attention coefficient according to the three-factor weights, and normalize the attention coefficient through the softmax function to obtain the attention weight of the spatiotemporal graph neural network; Based on the attention weights, edge weights based on a multi-head attention mechanism are obtained to generate GAT outputs.

10. A mine microseismic early warning device based on spatiotemporal graph neural network, characterized in that: The device comprises: a memory and a processor. When the processor runs the mine microseismic early warning program based on spatiotemporal graph neural network stored in the memory, the processor executes the method according to any one of claims 1 to 7.

Citation Information

Patent Citations

  • Intelligent early warning method for coal mine rock burst based on quantitative prediction of micro-seismic event

    CN113723595A

  • CGBoost algorithm for predicting susceptibility of co-seismic landslide

    CN116894523A

  • System and method for providing real-time prediction and mitigation of seismically-induced effects in complex systems

    US11143770B1

  • System and method for evaluating hydraulic system events and executing responses

    US20210026321A1