A well-ground joint enhanced microseismic positioning method, equipment, medium and product

Abstract

The present application discloses a well-ground joint enhanced microseismic positioning method, equipment, medium and product, which relates to the field of microseismic positioning. The method includes obtaining the first waveform data and the second waveform data of the microseismic event to be located in the target coal mine in a set period of time; the first waveform data is monitored by a ground station; the second waveform data is recorded by an optical fiber detector; the first waveform data and the second waveform data are pre-processed respectively to obtain the processed first waveform data and the processed second waveform data; according to the processed first waveform data and the processed second waveform data, the well-ground joint enhanced microseismic positioning model is used to determine the source position of the microseismic event to be located in the target coal mine; wherein the well-ground joint enhanced microseismic positioning model is obtained by training the initial model using a training data set; the initial model includes a graph convolutional network, a convolutional neural network and a prediction network connected in sequence. The present application improves the microseismic positioning accuracy of coal mines.

Description

Technical Field

[0001] The present application relates to the field of microseismic positioning, and in particular to a well-ground combined enhanced microseismic positioning method, equipment, medium and product. Background Art

[0002] In the process of coal mining, common dynamic disasters mainly include rock fracture, induced microseismicity and gas outburst accidents. These disasters are usually caused by factors such as stress changes in coal seams, water injection, gas release, etc., which may lead to tunnel collapse, coal dust explosion, equipment damage, and even casualties. Especially in high-pressure deep layers and complex geological structure areas, the frequency and risk of dynamic disasters increase significantly. Therefore, real-time monitoring and precise positioning of induced microseismicity are important links to ensure the safety of coal mining.

[0003] Microseismic monitoring in coal mining is one of the key tools for predicting and preventing dynamic disasters. By accurately locating microseismic events, it is possible to identify the location of rock fractures, monitor the effect of water injection, and assess the risk of induced microseisms in real time. This microseismic positioning technology helps operators identify potential danger areas at an early stage, providing an important basis for tunnel support and risk management, thereby ensuring the safety and stability of underground operations.

[0004] Traditional microseismic positioning methods, such as source scanning, double-difference positioning, and first-arrival wave inversion, usually rely on microseismic wave propagation time and waveform analysis, which are limited by data processing speed and error accumulation under complex geological conditions. Deep learning methods can significantly improve the speed and accuracy of microseismic positioning by automatically extracting spatial and temporal features. Deep learning models trained based on specific conditions of coal mining can better adapt to the complexity and nonlinear characteristics of rock formations, excel in real-time monitoring and early warning, and provide support for precise positioning and dynamic risk assessment.

[0005] At present, microseismic positioning technology in coal mining mainly relies on traditional travel time analysis and waveform matching methods, such as full waveform inversion, double difference positioning, and staggered correlation method. However, due to the complex geological conditions and heterogeneous coal seams in coal mines, traditional methods have obvious limitations in positioning accuracy. In addition, the introduction of joint well-ground monitoring systems has increased the diversity and complexity of data. Although more underground and surface monitoring stations provide rich data sources, too many monitoring points do not always bring about a linear improvement in positioning accuracy due to cumulative time errors and signal-to-noise ratio problems. Summary of the invention

[0006] The purpose of this application is to provide a well-ground combined enhanced microseismic positioning method, equipment, medium and product to improve the microseismic positioning accuracy of coal mines.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a well-ground joint enhanced microseismic positioning method, comprising:

[0009] Acquire the first waveform data and the second waveform data of the microseismic event to be located in the target coal mine in a set time period; the first waveform data is the microseismic waveform data monitored by the ground station; the second waveform data is the waveform data recorded by the optical fiber detector;

[0010] Preprocessing the first waveform data and the second waveform data respectively to obtain processed first waveform data and processed second waveform data; the preprocessing includes noise reduction processing, filtering processing and normalization processing;

[0011] According to the processed first waveform data and the processed second waveform data, the source position of the microseismic event to be located in the target coal mine is determined by using the joint well-ground enhanced microseismic positioning model; wherein the joint well-ground enhanced microseismic positioning model is obtained by training an initial model using a training data set; the training data set includes the first waveform data and the second waveform data of each microseismic event in the training coal mine in a set time period and the corresponding source position labels; the initial model includes a graph convolutional network, a convolutional neural network and a prediction network connected in sequence.

[0012] Optionally, preprocessing the first waveform data and the second waveform data respectively to obtain processed first waveform data and processed second waveform data specifically includes:

[0013] Performing noise reduction processing on the first waveform data and the second waveform data to obtain noise-reduced first waveform data and noise-reduced second waveform data;

[0014] Performing filtering processing on the first waveform data after noise reduction and the second waveform data after noise reduction to obtain filtered first waveform data and filtered second waveform data;

[0015] The filtered first waveform data and the filtered second waveform data are normalized to obtain processed first waveform data and processed second waveform data.

[0016] Optionally, the initial model is trained using the training data set to obtain a well-ground joint enhanced microseismic positioning model, specifically including:

[0017] Build a training dataset;

[0018] Preprocessing the training data set to obtain a processed training data set;

[0019] Inputting the processed first waveform data and the processed second waveform data of each microseismic event of the training coal mine in a set time period into the current initial model to obtain the predicted source position of the microseismic event of the training coal mine;

[0020] According to the predicted earthquake source location and the corresponding earthquake source location label, a loss function value is determined using a total loss function; the total loss function is determined based on a mean square error loss function and an L2 regularization loss function;

[0021] Determine whether the training end condition is met; the training end condition is that the maximum number of iterations is reached or the loss function value is less than a set value;

[0022] If yes, the current initial model is used as the well-ground joint enhanced microseismic positioning model;

[0023] If not, adjust the model parameters of the current initial model according to the loss function value, and return to "input the processed first waveform data and the processed second waveform data of each microseismic event in the training coal mine in the set time period into the current initial model to obtain the predicted epicenter position of the microseismic event in the training coal mine."

[0024] Optionally, the total loss function is:

[0025] ;

[0026] in, is the total number of microseismic events, is the source location label, To predict the location of earthquake source; is the regularization parameter; is the model weight.

[0027] Optionally, the graph convolutional network includes a first graph convolutional layer, a second graph convolutional layer and a third graph convolutional layer connected in sequence.

[0028] Optionally, the convolutional neural network includes a first convolution block, a second convolution block and a third convolution block connected in sequence; the first convolution block includes a first convolution layer and a first maximum pooling layer connected in sequence; the second convolution block includes a second convolution layer and a second maximum pooling layer connected in sequence; the third convolution block includes a third convolution layer and a global average pooling layer connected in sequence.

[0029] Optionally, the prediction network includes a first fully connected layer and a second fully connected layer connected in sequence.

[0030] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described well-ground joint enhanced microseismic positioning methods.

[0031] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described well-ground joint enhanced microseismic positioning methods.

[0032] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned well-ground joint enhanced microseismic positioning methods.

[0033] According to the specific embodiments provided in this application, this application has the following technical effects:

[0034] The present application provides a method, device, medium and product for joint well-ground enhanced microseismic positioning, which obtains the first waveform data and the second waveform data of the microseismic event to be located in the target coal mine in a set time period; the first waveform data is the microseismic waveform data monitored by the ground station; the second waveform data is the waveform data recorded by the optical fiber detector; the first waveform data and the second waveform data are preprocessed respectively to obtain the processed first waveform data and the processed second waveform data; according to the processed first waveform data and the processed second waveform data, the well-ground joint enhanced microseismic positioning model is used to determine the source position of the microseismic event to be located in the target coal mine; wherein the well-ground joint enhanced microseismic positioning model is obtained by training the initial model using the training data set; the initial model includes a graph convolutional network, a convolutional neural network and a prediction network connected in sequence. The present application combines the underground optical fiber and the three-component surface station data, and uses a deep learning model for joint well-ground microseismic positioning. The deep learning model combines the convolutional neural network, the graph convolutional network and the prediction network, which can realize the accurate prediction of the location of the microseismic event, and has extremely high computational efficiency and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 A schematic diagram of a flow chart of a well-ground joint enhanced microseismic positioning method provided in one embodiment of the present application;

[0037] Figure 2 This is a schematic diagram of the implementation process of the well-ground combined enhanced microseismic positioning method of the present application;

[0038] Figure 3 An event station distribution diagram provided in an embodiment of the present application;

[0039] Figure 4 A schematic diagram of the location of a well-ground joint monitoring system provided in one embodiment of the present application;

[0040] Figure 5 A schematic diagram of time distribution provided in an embodiment of the present application; wherein (a) is a schematic diagram of the original event distribution of the training data set; (b) is a schematic diagram of the oversampled event distribution of the training data set;

[0041] Figure 6 This is a schematic diagram of the TSLN model architecture;

[0042] Figure 7 This is a graph showing how the accuracy and loss function change with the number of training rounds during model training;

[0043] Figure 8 This is a schematic diagram of the positioning error analysis of the test set;

[0044] Fig. 9 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0047] This application combines underground optical fiber and surface three-component station data and uses deep learning to perform joint well-ground microseismic positioning, which can more efficiently process multimodal data, overcome the shortcomings of traditional methods, and provide a new technical solution for microseismic monitoring and disaster warning in coal mining.

[0048] In an exemplary embodiment, Figure 1 As shown, a well-ground combined enhanced microseismic positioning method is provided, comprising the following steps:

[0049] S1: Acquire the first waveform data and the second waveform data of the microseismic event to be located in the target coal mine in a set time period; the first waveform data is the microseismic waveform data monitored by the ground station; the second waveform data is the waveform data recorded by the optical fiber detector.

[0050] S2: Preprocessing the first waveform data and the second waveform data respectively to obtain processed first waveform data and processed second waveform data; the preprocessing includes noise reduction processing, filtering processing and normalization processing.

[0051] As an optional implementation, S2 specifically includes:

[0052] S21: performing noise reduction processing on the first waveform data and the second waveform data to obtain noise-reduced first waveform data and noise-reduced second waveform data.

[0053] S22: filtering the first waveform data after noise reduction and the second waveform data after noise reduction to obtain filtered first waveform data and filtered second waveform data.

[0054] S23: performing normalization processing on the filtered first waveform data and the filtered second waveform data to obtain processed first waveform data and processed second waveform data.

[0055] S3: According to the processed first waveform data and the processed second waveform data, the source position of the microseismic event to be located in the target coal mine is determined by using the joint well-ground enhanced microseismic positioning model; wherein the joint well-ground enhanced microseismic positioning model is obtained by training an initial model using a training data set; the training data set includes the first waveform data and the second waveform data of each microseismic event in the training coal mine in a set time period and the corresponding source position label; the initial model includes a graph convolutional network, a convolutional neural network and a prediction network connected in sequence.

[0056] As an optional implementation, the graph convolution network includes a first graph convolution layer, a second graph convolution layer, and a third graph convolution layer connected in sequence. The convolution neural network includes a first convolution block, a second convolution block, and a third convolution block connected in sequence; the first convolution block includes a first convolution layer and a first maximum pooling layer connected in sequence; the second convolution block includes a second convolution layer and a second maximum pooling layer connected in sequence; the third convolution block includes a third convolution layer and a global average pooling layer connected in sequence. The prediction network includes a first fully connected layer and a second fully connected layer connected in sequence.

[0057] This application develops a well-ground joint enhanced microseismic positioning model called TSLN (Tri-Level Seismic Localization Network) to improve the positioning accuracy and efficiency of microseismic events in coal mining. The model combines a graph convolutional network (GCN), a convolutional neural network (CNN) and a prediction network (PN), and through a joint monitoring strategy of underground shaft fiber optic sensors and surface three-component stations, comprehensively extracts spatial features, realizes multimodal data fusion in vertical and horizontal directions, and significantly improves positioning accuracy. GCN models the correlation of spatial nodes, CNN is used to capture the local dependency features of GCN output, and PN consists of several fully connected layers, focusing on global nonlinear combination and classification of high-level features for predicting the location of earthquake sources. For surface stations, a pentagonal completeness envelope with a clearance angle less than 90° is selected to ensure the coverage and accuracy of surface monitoring; the underground vertical fiber optic well (such as Well①) is equipped with 20 fiber optic detectors with a channel spacing of 10 meters to enhance the depth constraint capability and facilitate accurate capture of the source location of deep microseismic events. Based on the pre-trained model of training samples, TSLN is significantly superior to traditional microseismic positioning algorithms in positioning accuracy and computational efficiency. It can locate a new microseismic event within 0.1 seconds, with an absolute error of less than 40 meters, and has the potential for meter-level accuracy with sufficient training data.

[0058] Traditional methods for locating microseismic sites in mining areas usually use grid search or full waveform inversion, using data from all monitoring stations for positioning. However, previous studies have shown that selecting at least 6 stations around the earthquake source for data analysis can achieve better positioning results. Based on this concept, the TSLN model selects multimodal data from joint well-ground monitoring for specific areas for model training, and introduces the GCN-CNN-PN architecture. Specifically, GCN captures the correlation of spatial nodes through graph convolution, extracts the global spatial relationship between ground stations and fiber optic detectors, and CNN extracts local features of GCN output through 1D convolution. PN is a feedforward network that fuses the features extracted by GCN and CNN to complete the final positioning prediction of the earthquake source coordinates, forming an efficient deep learning microseismic positioning solution.

[0059] A large number of experiments have verified the superior performance of the TSLN model in microseismic location tasks, opening up a new path for the application of deep learning technology in microseismic monitoring in coal mining. Figure 2 As shown in the figure, the method consists of three steps: dataset creation, model construction and training, and earthquake source location prediction.

[0060] As an optional implementation method, the initial model is trained using the training data set to obtain a well-ground joint enhanced microseismic positioning model, which specifically includes:

[0061] Step 1: Build a training dataset.

[0062] In actual applications, the dataset creation process is as follows:

[0063] A mine in China is conducting long-term coal mining, which poses a risk of inducing microseisms. Therefore, a microseismic intensive station monitoring system has been deployed on the ground in the mining area. At the same time, three vertical shafts were drilled in the area to identify the thickness and location of the coal seams, and fiber optic detectors were laid out for vertical microseismic profiling (Vertical Seismic Profiling, VSP). The fiber optic detectors can also record microseismic events. The monitoring period was from January 15, 2021 to December 15, 2021, with a total of 1,704 events, including 312 microseismic events below magnitude 0, 1,228 microseismic events of magnitude 0 to 1, 151 microseismic events of magnitude 1 to 2, and 13 microseismic events of magnitude > 2. The maximum magnitude was 2.88, which were basically imperceptible microseisms. The event stations are distributed as follows: Figure 3 shown.

[0064] Considering that the densely populated stations on the surface have strong constraints on the horizontal direction and weak constraints on the depth direction, while the fiber optic detectors in the vertical shaft have strong constraints on the depth direction and weak constraints on the horizontal direction, it is considered to build a joint monitoring system for wells and ground. The fiber optic vertical shaft Well① (20 channels, channel spacing of 10 meters) with the most effective fiber optic detectors in the mining area is selected as the center of the monitoring system. The surface is arranged according to the completeness of the stations, and 5 stations near the center of Well① are selected so that the station gap angle is less than 90° to form a pentagonal envelope. The joint monitoring system for wells and ground is as follows: Figure 4 shown.

[0065] The data set was constructed from 1704 events monitored by the microseismic dense array monitoring system, all of which were three-component data with a sampling rate of 100Hz. The waveform data of 5 seconds before and 5 seconds after the earthquake was intercepted, with a total of 5000 sampling points. The input data included 5000 points of microseismic waveform data from 5 ground stations for each microseismic event and 5000 points of waveform data recorded by 20 fiber optic detectors in the corresponding period. Initially, the waveform data was uniformly denoised, filtered and normalized, and then the data set was shuffled and divided into a training data set and a test data set in a ratio of 8:2.

[0066] Figure 5(a) in the figure is the original distribution of microseismic events in the training data set. It can be found that the data distribution is unbalanced. In the partition of the working face, the maximum number of events in some areas can reach 308, and the minimum is 0. Training with this training data set will cause the model to predict the majority class and ignore the minority class samples, and the generalization ability will be weakened. Therefore, after the data set is shuffled, it is first divided into a training data set and a test data set, and then the training data set is oversampled. Based on the maximum number of events in the area of ​​308, the samples of some areas with less than 308 events are copied, except for the areas with no events originally, such as Figure 5 As shown in (b), after oversampling the training set, the total number of samples reached 7084 events.

[0067] The constructed training data set consists of 7084 events, 5000 amplitude data points from 5 ground stations and 5000 amplitude data points recorded by 20 fiber optic detectors in the corresponding period for each microseismic event, and the corresponding label set is the source location (X, Y, Z).

[0068] Step 2: Preprocess the training data set to obtain a processed training data set.

[0069] Step 3: Input the processed first waveform data and the processed second waveform data of each microseismic event of the training coal mine in a set time period into the current initial model to obtain the predicted source position of the microseismic event of the training coal mine.

[0070] Step 4: According to the predicted source location and the corresponding source location label, the loss function value is determined using the total loss function; the total loss function is determined based on the mean square error loss function and the L2 regularization loss function.

[0071] The total loss function is:

[0072] .

[0073] in, is the total number of microseismic events, is the source location label, To predict the location of earthquake source; is the regularization parameter; is the weight of the model.

[0074] Step 5: Determine whether the training end condition is met; the training end condition is that the maximum number of iterations is reached or the loss function value is less than a set value.

[0075] Step 6: If yes, the current initial model is used as the well-ground joint enhanced microseismic positioning model.

[0076] Step 7: If not, adjust the model parameters of the current initial model according to the loss function value, and return to "input the processed first waveform data and the processed second waveform data of each microseismic event in the training coal mine in the set time period into the current initial model to obtain the predicted epicenter position of the microseismic event in the training coal mine."

[0077] After the data set is constructed, the model is constructed and trained.

[0078] The model architecture of TSLN is as follows Figure 6 As shown in the figure, the earthquake source location is obtained mainly by using the graph convolutional network (GCN), convolutional neural network (CNN) and then the fully connected layer.

[0079] The input layer of the model is the ground station three-dimensional data (5, 3, 5000) and the optical fiber single component data (20, 5000). The ground station three-dimensional data is averaged in the third dimension, and the three-component data of each station is compressed into a single channel to obtain the average amplitude value of each station. The new station data shape is (5, 5000) and is merged with the optical fiber detector data in the node dimension to form a feature matrix of (25, 5000).

[0080] The first layer of the model is the graph convolutional network layer GCN. The GCN part first needs to build a graph structure according to the location of the station. Each station corresponds to a node in the graph, and the weight of the edge between the nodes corresponds to the spatial straight-line distance between the stations. A 25×25 adjacency matrix is ​​constructed based on the nodes and edges. The waveform length corresponding to each node is 5000, so the size of the feature matrix is ​​25×5000. The GCN part can fully extract the spatial relationship between the stations. After the graph structure is built, the amplitude data is merged into three GCN convolutional layers. Each layer has special parameters to extract and abstract spatial features from the data. The first layer has 1024 neurons, the second layer has 256 neurons, and the third layer has 64 neurons. After being processed by the graph convolutional network, the input data size changes from 25×5000 to 25×64. The formula is:

[0081] (1)

[0082] in, For the l +1 layer node feature matrix; is the adjacency matrix, is the degree matrix, It is The node feature matrix of the layer, It is The weight matrix of the layer, is the activation function (such as ReLU).

[0083] The input data shape of the graph convolutional network is: num_nodes=25, num_features=5000, consisting of 5 ground stations and 20 fiber detectors, with 25 nodes and an adjacency matrix size of (25, 25).

[0084] Graph convolution layer 1 (first graph convolution layer): 1024 graph convolution kernels, batch normalization (BN) and ReLU activation, output shape (25, 1024).

[0085] Graph convolution layer 2 (second graph convolution layer): 256 graph convolution kernels, batch normalization (BN) and ReLU activation, output shape (25, 256).

[0086] Graph convolution layer 3 (the third graph convolution layer): 64 graph convolution kernels, using batch normalization (BN) and ReLU activation, and output shape (25, 64).

[0087] The output data shape of the graph convolutional network is: num_nodes=25, num_features=64.

[0088] The convolutional neural network (CNN) extracts local feature patterns of the time series from the spatial node features provided by GCN, captures the time-dependent characteristics of the signal, and uses a 1D convolution layer to identify the temporal local patterns in the vibration signal. The temporal local features of the vibration signal are extracted and combined with the spatial features of GCN to form a more discriminative feature representation. The convolution operation is defined as:

[0089] (2)

[0090] in, represents the convolution kernel, represents the input feature map, is the output coordinate, is the size of the convolution kernel.

[0091] The input data shape of the convolutional neural network is: channels=25, sequence_length=64.

[0092] Convolutional block 1 (first convolutional block): (25, 64) → (64, 64) → (64, 32).

[0093] The first convolutional layer: 64 1D convolution kernels of length 3, padded with 1 on both sides of the sequence, using batch normalization (BN) and ReLU activation, and the output shape is (64, 64).

[0094] First max pooling layer: 1D max pooling kernel of length 2, stride 2, and output shape (64, 32).

[0095] Convolution block 2 (second convolution block): (64, 32) → (128, 32) → (128, 16).

[0096] The second convolutional layer: 128 1D convolution kernels of length 3, padded with 1 on both sides of the sequence, using batch normalization (BN) and ReLU activation, and the output shape is (128, 32).

[0097] Second max pooling layer: 1D max pooling kernel of length 2, stride 2, and output shape (128, 16).

[0098] Convolution block 3 (the third convolution block):

[0099] The third convolutional layer: 256 1D convolution kernels of length 3, padded with 1 on both sides of the sequence, using batch normalization (BN) and ReLU activation, and the output shape is (256, 16).

[0100] Global average pooling layer: compresses the sequence length to 1 and the output shape is (256, 1).

[0101] The prediction network consists of two fully connected layers, which are used to fuse the high-dimensional global features output by CNN to generate the final microseismic location results. The formula is:

[0102] (3)

[0103] in, is the weight matrix, is the high-dimensional global feature output by CNN, is the bias term.

[0104] The input of the prediction network needs to be flattened first: (256, 1) → (256).

[0105] Fully connected layer 1 (first fully connected layer): input 256-dimensional features and output 64-dimensional features;

[0106] Fully connected layer 2 (second fully connected layer): input 64-dimensional features and output 3-dimensional features, corresponding to the (x, y, z) coordinates of the earthquake source location.

[0107] Loss function of the TSLN model: The loss function of the model uses a combination of mean square error and L2 regularization, and its formula is:

[0108] (4)

[0109] The mean squared error (MSE) is used to measure the difference between the predicted earthquake source location and the true value. By minimizing MSE, the model can optimize the positioning accuracy.

[0110] (5)

[0111] in, is the number of samples (total number of microseismic events), is the true earthquake source location, To predict the location of the earthquake source.

[0112] L2 regularization suppresses overfitting of the model by introducing the sum of squares of model parameters into the loss function. It penalizes large weight values, prompting the model to learn smoother functions, and helps improve the generalization ability of the model and avoid overfitting on the training data. The formula is:

[0113] (6)

[0114] The introduction of L2 regularization helps the model avoid overfitting. Since deep learning models usually have a large number of trainable parameters, they are prone to perform too well on training data but poorly on test data. Through L2 regularization, the model is forced to learn more general features, reducing the risk of overfitting.

[0115] In the training of the TSLN model, the Adam optimizer is used for optimization, and the initial learning rate is set to 0.0001 to ensure stable convergence under complex data structures. During the training process, mini-batch gradient descent with a batch size of 32 is used, combined with an adaptive learning rate adjustment strategy - ReduceLROnPlateau. This callback mechanism will automatically halve the learning rate when the validation loss has not decreased significantly for 5 consecutive cycles, so that the model can converge to the optimal solution more finely. At the same time, in order to effectively prevent the overfitting problem, the Early Stopping mechanism is introduced. When the validation loss does not improve within 20 cycles, the training process will automatically stop and restore to the optimal weight, thereby ensuring that the model retains the best performance while avoiding overfitting. The total training cycle is capped at 200, and whenever the model's performance on the test set exceeds all previous cycles, the current model weight will be automatically saved as the best model. This training strategy effectively improves the positioning accuracy and stability of the TSLN model, ensuring the reliability and robustness of the model in practical applications.

[0116] Figure 7 The changing trends of loss function and accuracy during model training are shown in Figure 2. The blue curve represents training loss, the green dotted line represents validation loss, the red dotted line represents training accuracy, and the purple dotted line represents validation accuracy.

[0117] from Figure 7It can be observed that the training loss of the model drops rapidly in the early training stage, which indicates that the model can learn from the data quickly and reduce the error in the early stage. Similarly, the validation loss also drops rapidly in the early stage, but after about 100 cycles, the validation loss curve gradually flattens out, and the loss value fluctuates slightly at 0.2, indicating that the model may have approached the limit of its generalization ability. At the same time, the training accuracy (red dotted line) rises rapidly in the early stages of training and gradually stabilizes in the later stages. The validation accuracy (purple dotted line) also shows a similar trend, approaching a stable state after about 100 cycles. In the end, the best performing version of the model achieved an accuracy of 0.91 in the 150th cycle, reaching a relatively ideal state.

[0118] In order to evaluate the overall performance and estimation error of the model, the errors of all prediction results in the test set were counted, and the error judgment criteria were defined, including horizontal error (HE), depth error (DE), and absolute error (AE):

[0119] (7)

[0120] (8)

[0121] (9)

[0122] Among them, (x pred ,y pred , z pred ) is the predicted earthquake source location, (x true ,y true , z true ) is the true earthquake source location.

[0123] The localization error of the test set is Figure 8 As shown in the figure, the horizontal error is low in most cases, mainly concentrated between 0 and 20 meters, indicating that the model has good positioning accuracy in the horizontal direction. However, there are some large error values, which may be caused by a few extreme events or noisy data. The depth errors are almost all concentrated between 0 and 10 meters, and a few errors are beyond this range. Compared with the horizontal error, the depth error has a higher concentration, showing good depth prediction ability. The absolute error values ​​are mostly concentrated between 0 and 40 meters, with some large errors, but the overall error level is still low, showing the stability of the model in overall positioning.

[0124] After the model training is completed, the earthquake source location is predicted.

[0125] After the training, the best model was applied to the prediction of microseismic events. Eight blasting events in the coal mine were selected, and the specific information is shown in Table 1.

[0126] Table 1 Statistics of explosion incidents in a coal mine

[0127]

[0128] The magnitude of the earthquake was between 1.86 and 2.21, and the waveform data quality of these events was high. After the waveforms of all events were denoised, filtered, and amplitude normalized, the source location and timestamp were predicted. For these 8 events, only ground stations (using the hypoDD algorithm), only optical fibers (using the grid search algorithm), and the joint well-ground monitoring system of the present application method were used for positioning, and the positioning error was calculated with the known blasting location. The calculation results are shown in Table 2.

[0129] Table 2 Positioning error analysis table

[0130]

[0131] As can be seen from Table 2, the TSLN model has both accuracy and speed advantages in positioning. The ground station has a significant constraint effect on the horizontal direction, and the vertical shaft fiber has higher accuracy in the depth direction. TSLN combines the advantages of these two networks, and the errors in the horizontal and depth directions are smaller than those of a single network. At the same time, TSLN has extremely high computational efficiency, and a single event prediction only takes about 0.1 seconds, which is significantly better than the hypoDD algorithm used by the ground station (about 1 second) and the grid search method of the fiber network (about 1200 seconds).

[0132] The well-ground joint enhanced microseismic positioning model of this application is specifically used for the positioning of microseismic events induced by coal mining. The model combines convolutional neural networks (CNN), graph convolutional networks (GCN) and prediction networks (PN), which can achieve accurate prediction of the location of microseismic events and has extremely high computational efficiency and prediction accuracy. TSLN takes only 0.1 seconds to predict a single event, and the positioning error is controlled within 40 meters. With sufficient training samples, TSLN is expected to achieve fast and high-precision positioning, and the positioning error can be further reduced.

[0133] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above-mentioned well-ground joint enhanced microseismic positioning method when executing the computer program.

[0134] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the above-mentioned well-ground joint enhanced microseismic positioning method is implemented.

[0135] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements the above-mentioned well-ground joint enhanced microseismic positioning method when executed by a processor.

[0136] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a well-ground joint enhanced microseismic positioning method is implemented.

[0137] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0138] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0139] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0140] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0141] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0142] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A well-ground joint enhanced microseismic positioning method, characterized in that: include: Acquire the first waveform data and the second waveform data of the microseismic event to be located in the target coal mine in a set time period; The first waveform data is microseismic waveform data monitored by a ground station; The second waveform data is waveform data recorded by using a fiber optic detector; Preprocessing the first waveform data and the second waveform data respectively to obtain processed first waveform data and processed second waveform data; the preprocessing includes noise reduction processing and normalization processing; According to the processed first waveform data and the processed second waveform data, the source position of the microseismic event to be located in the target coal mine is determined by using the joint well-ground enhanced microseismic positioning model; wherein the joint well-ground enhanced microseismic positioning model is obtained by training an initial model using a training data set; the training data set includes the first waveform data and the second waveform data of each microseismic event in the training coal mine in a set time period and the corresponding source position labels; the initial model includes a graph convolutional network, a convolutional neural network and a prediction network connected in sequence.

2. The well-ground combined enhanced microseismic positioning method according to claim 1, characterized in that: The initial model is trained using the training data set to obtain a well-ground joint enhanced microseismic positioning model, which specifically includes: Build a training dataset; Preprocessing the training data set to obtain a processed training data set; Inputting the processed first waveform data and the processed second waveform data of each microseismic event of the training coal mine in a set time period into the current initial model to obtain the predicted source position of the microseismic event of the training coal mine; According to the predicted earthquake source location and the corresponding earthquake source location label, a loss function value is determined using a total loss function; the total loss function is determined based on a mean square error loss function and an L2 regularization loss function; Determine whether the training end condition is met; the training end condition is that the maximum number of iterations is reached or the loss function value is less than a set value; If yes, the current initial model is used as the well-ground joint enhanced microseismic positioning model; If not, adjust the model parameters of the current initial model according to the loss function value, and return "input the processed first waveform data and the processed second waveform data of each microseismic event in the training coal mine in the set time period into the current initial model to obtain the predicted epicenter position of the microseismic event in the training coal mine." 3. The well-ground combined enhanced microseismic positioning method according to claim 2, characterized in that: The total loss function is: ； in, is the total number of microseismic events, is the source location label, To predict the location of earthquake source; is the regularization parameter; is the model weight.

4. The well-ground combined enhanced microseismic positioning method according to claim 1, characterized in that: The graph convolutional network includes a first graph convolutional layer, a second graph convolutional layer and a third graph convolutional layer which are connected in sequence.

5. The well-ground combined enhanced microseismic positioning method according to claim 1, characterized in that: The convolutional neural network includes a first convolution block, a second convolution block and a third convolution block connected in sequence; the first convolution block includes a first convolution layer and a first maximum pooling layer connected in sequence; the second convolution block includes a second convolution layer and a second maximum pooling layer connected in sequence; the third convolution block includes a third convolution layer and a global average pooling layer connected in sequence.

6. The well-ground combined enhanced microseismic positioning method according to claim 1, characterized in that: The prediction network includes a first fully connected layer and a second fully connected layer connected in sequence.

7. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the well-ground joint enhanced microseismic positioning method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the well-ground combined enhanced microseismic positioning method described in any one of claims 1 to 6 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the well-ground combined enhanced microseismic positioning method described in any one of claims 1 to 6 is implemented.

Images