A low signal-to-noise ratio seismic detection model based on a deep residual shrinkage network

Abstract

The application discloses a low signal-to-noise ratio seismic detection model based on a deep residual shrinkage network, and in view of the difference of noise in seismic records collected by different stations or geophones, a deep residual shrinkage network is constructed by introducing an attention mechanism on the basis of designing a soft threshold function, and is used for detecting seismic events in low signal-to-noise ratio seismic records. In the designed network, the soft threshold function is inserted into the deep network structure as a nonlinear transformation layer, so as to effectively filter the influence of noise-related features. Meanwhile, the attention mechanism is introduced into the network, and an adaptive slope module is added, so that the optimal threshold for seismic signal denoising is automatically obtained, and the best processing of different signals and different degrees of noise is realized. Through training, the deep residual shrinkage network can adaptively determine the denoising threshold, so that each seismic signal has its own threshold set, and the purpose of accurate detection of seismic events under strong background noise is achieved, and the accuracy of seismic event detection is improved.

Description

Technical Field

[0001] This invention belongs to the field of earthquake detection technology, specifically relating to a low signal-to-noise ratio earthquake detection model based on a deep residual shrinkage network. Background Technology

[0002] Seismic event detection refers to the identification of seismic events from continuous seismic records, and it is a core issue in seismological observation. Aftershock monitoring and rapid location after major earthquakes, tomographic imaging of the Earth's internal velocity structure, and the compilation of seismic catalogs all rely on seismic event detection. With the continuous expansion of seismic networks and mobile arrays, and the exponential increase in density, seismic observation data is growing explosively. Relying on human-computer interaction to detect seismic events from massive seismic records is no longer sufficient to meet the demands of intelligent seismic data processing in the era of big data. Therefore, how to accurately and automatically detect seismic events from seismic records has become a pressing problem that needs to be solved in seismic data processing and application.

[0003] In existing technologies, deep learning-based methods have achieved certain results in seismic event detection, but they require high-quality seismic data. For high signal-to-noise ratio (SNR) seismic records, they generally yield good detection results, but for low SNR seismic records (such as microseismic events, whose energy is extremely weak, even at the level of noise), the reliability of the detection results may be questionable. Due to the complex and variable nature of the seismic observation environment, the acquired seismic records contain various types of noise, which can severely affect the accuracy of seismic event detection.

[0004] Therefore, seismic records are typically denoised before detection. Common seismic signal denoising methods include empirical mode decomposition (EMD) and wavelet thresholding. The main principle of these methods is to transform the original signal to a less important near-zero region, and then apply a threshold function to convert near-zero features to zero, thus achieving denoising. Usually, the denoising threshold for these methods is fixed. However, studies have found that the noise levels in seismic records acquired by different stations or detectors vary significantly, leading to poor denoising results with these methods. This is especially true for seismic records with low signal-to-noise ratios. For example, for microseismic signals generated by hydraulic fracturing, the complex monitoring environment and strong background noise interference cause weak microseismic events to be submerged in the background noise. Therefore, inappropriate denoising processing can lead to poor detection results, especially for microseismic events.

[0005] Threshold function design is a crucial step in signal denoising. To improve denoising performance, soft thresholding functions have been introduced into many seismic signal denoising methods. However, setting a suitable soft threshold function remains highly challenging, requiring extensive expertise in seismic signal processing. Furthermore, the noise levels and intensities of seismic records acquired from different stations or detectors vary significantly, leading to potentially different optimal threshold functions for different signals. This further complicates the design of soft threshold functions. Summary of the Invention

[0006] The purpose of this invention is to provide a low signal-to-noise ratio (SNR) earthquake detection model based on a deep residual contraction network. By designing a soft threshold function and introducing an attention mechanism, a deep residual contraction network is constructed for detecting earthquake events in low SNR seismic records, effectively improving the accuracy of earthquake event detection. To achieve the above objective, the technical solution adopted by this invention is as follows: A low signal-to-noise ratio seismic detection model based on a deep residual shrinkage network includes the following steps: A. Construct a deep residual shrinkage network by stacking a certain number of residual shrinkage units based on adaptive slope modules (RSBU-ASB), input layer, convolutional layer, batch normalization (BN), ReLU activation function, global average pooling (GAP) and cross-entropy error function together to construct a complete deep residual shrinkage network. A1. A soft thresholding function is inserted as a nonlinear transformation layer into a deep residual shrinkage network structure to filter out the influence of noise-related features; A2. An attention mechanism is introduced into the network, and an adaptive slope module is added to automatically obtain the optimal threshold for denoising seismic signals, thereby achieving optimal processing of different signals and different levels of noise. B. Establish an earthquake detection model to accurately obtain the detection results of earthquake events.

[0007] Furthermore, in step A, a seismic signal of size C×W×1 is input through the input layer, where C is the number of channels, W is the width, and 1 is the height of the seismic signal (since the input is a three-dimensional seismic waveform signal, the height is always 1); then, a convolutional layer is used to extract the features of each channel of the signal. In order to prevent overfitting and increase the nonlinear expressive power of the model, BN and ReLU activation functions are added before the convolution operation.

[0008] Furthermore, in step A, the convolutional layer extracts features from the input object and multiplies the output features of the previous layer with the convolutional kernel according to certain rules. , Formula 1; in To output the j-th feature of the l-th layer in the feature map, For the i-th feature in layer l-1 of the input feature map, Let the kernel size be the convolution kernel size between the j-th feature of layer l and the i-th feature of layer l-1. For the corresponding bias, * indicates the convolution operation, M j This indicates that it is used to calculate the j-th feature.

[0009] Furthermore, in step A, after extracting the features of the input object through the convolutional layer, the extracted data is then normalized through batch normalization to improve the convergence speed and generalization ability of the convolutional neural network, and the stretching parameter γ and the offset parameter are introduced. To recover the feature distribution learned by the network layers; in Formula 2; Formula 3; Formula 4; Formula 5; In the formula, m is the batch size of the sample, and x i and y i These are the i-th input feature and the output feature in the sample, respectively.

[0010] Furthermore, after performing the convolution operation, the threshold of the soft thresholding function is estimated using a soft thresholding module, and an adaptive slope module (ASB) is added to form a residual shrinking unit (RSBU-ASB) based on the adaptive slope module. The most suitable slope of the signal is automatically inferred using an attention mechanism to further correct the threshold.

[0011] Furthermore, when using the soft thresholding module for threshold estimation, a one-dimensional vector is first obtained by applying a gap to the absolute value of the feature map, and then the one-dimensional vector is passed to two fully connected (FC) layers to obtain the scale parameter α. c Finally, the sigmoid function is used to scale the scale parameter to the range (0, 1). The formula for calculating the scale parameter is: Formula Six Among them, z c For the characteristics of neurons in layer c, τ c The threshold is obtained by multiplying the scale parameter by the absolute value of the one-dimensional feature vector, i.e. Formula 7; Where i, j, and c are the width, height, and channel of the feature map x, respectively.

[0012] Furthermore, in the adaptive slope module, the most suitable slope for the signal is automatically inferred using an attention mechanism, and its output is: Formula 8; Where β is the adaptive slope factor, q c These are the characteristics of neurons in layer c; Finally, the output of the soft thresholding module is corrected based on the adaptive slope factor β to obtain the optimal soft threshold, i.e.: Form Nine; Based on the obtained optimal soft threshold, features with absolute values ​​less than the threshold are deleted, while features with absolute values ​​greater than the threshold are moved closer to 0, in order to achieve the purpose of shrinkage.

[0013] Furthermore, in step B, the established earthquake detection model includes a training set, a validation set, and a test set. The samples are randomly allocated to the established training set, validation set, and test set in a ratio of 8:1:1, respectively, for model training, model hyperparameter adjustment, and testing of the model's earthquake event detection performance.

[0014] Furthermore, multiple iterations of training are set up in the training set. Through repeated training and optimization experiments, the error of the training results is minimized, the model converges, and the optimal model is obtained.

[0015] Furthermore, after the model has been trained on the training set, the established earthquake detection model is validated using samples from the test set.

[0016] Compared with existing technologies, the advantages of this invention are as follows: This invention establishes an earthquake detection model by constructing a deep residual shrinkage network. In this model, a soft threshold function is inserted as a nonlinear transform layer into the deep network structure to effectively filter the influence of noise-related features. Simultaneously, an attention mechanism is introduced into the network, and an ASB (Automatic Seismic Block) is added to automatically obtain the optimal threshold for seismic signal denoising, achieving optimal processing of different signals and different levels of noise, thereby achieving the goal of accurate detection of earthquake events under strong background noise. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is the flowchart of the low signal-to-noise ratio earthquake detection model based on deep residual shrinkage network provided by the present invention.

[0019] Figure 2This is a graph of earthquake events and noise signal data from one of the data sources for the low signal-to-noise ratio earthquake detection model based on a deep residual shrinkage network provided by this invention.

[0020] Figure 3 This is a graph of earthquake events and noise signal data from the microseismic monitoring dataset of data source two, which is the low signal-to-noise ratio earthquake detection model based on deep residual shrinkage network provided by this invention.

[0021] Figure 4 This is a graph showing the accuracy and loss function of the model training process for the low signal-to-noise ratio earthquake detection model based on a deep residual shrinkage network provided by this invention.

[0022] Figure 5 It is the confusion matrix of the STEAD test set.

[0023] Figure 6 This is the ROC curve of the STEAD test set.

[0024] Figure 7 It is the confusion matrix of the microseismic monitoring data test set.

[0025] Figure 8 This is the ROC curve of the microseismic monitoring data test set. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0028] In the description of this invention, it should be noted that if terms such as "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," or "outer" are used to indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed, they are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0029] Furthermore, the terms "first," "second," and "third" are used only for distinguishing descriptions and should not be interpreted as indicating or implying relative importance.

[0030] Furthermore, the use of terms such as "horizontal," "vertical," and "suspended" does not imply that the component must be absolutely horizontal or suspended, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal relative to "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0031] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0032] It should be noted that, where there is no conflict, the features in the embodiments of the present invention can be combined with each other.

[0033] like Figure 1 As shown, a low signal-to-noise ratio seismic detection model based on a deep residual shrinkage network includes the following steps: A. Construct a deep residual shrinkage network by stacking a certain number of residual shrinkage units based on adaptive slope modules (RSBU-ASB), input layer, convolutional layer, batch normalization (BN), ReLU activation function, global average pooling (GAP) and cross-entropy error function together to construct a complete deep residual shrinkage network. First, a seismic signal of size C×W×1 is input through the input layer, where C is the number of channels, W is the width, and 1 is the height of the seismic signal (since the input is a three-dimensional seismic waveform signal, the height is always 1). Then, a convolutional layer is used to extract the features of each channel of the signal. In order to prevent overfitting and increase the nonlinear expressive power of the model, BN and ReLU activation functions are added before the convolution operation.

[0034] The convolutional layer extracts features from the input object, maps the output features of the previous layer to the convolutional kernel, and multiplies them according to certain rules. , Formula 1; in To output the j-th feature of the l-th layer in the feature map, For the i-th feature in layer l-1 of the input feature map, Let the kernel size be the convolution kernel size between the j-th feature of layer l and the i-th feature of layer l-1. For the corresponding bias, * indicates the convolution operation, M j This indicates that it is used to calculate the j-th feature.

[0035] After extracting features of the input object through convolutional layers, batch normalization is then applied to normalize the extracted data, improving the convergence speed and generalization ability of the convolutional neural network. A stretching parameter γ and an offset parameter are also introduced. To recover the feature distribution learned by the network layers; in Formula 2; Formula 3; Formula 4; Formula 5; In the formula, m is the batch size of the sample, and x i and y i These are the i-th input feature and the output feature in the sample, respectively.

[0036] A1. A soft thresholding function is inserted as a nonlinear transformation layer into a deep residual shrinkage network structure to filter out the influence of noise-related features.

[0037] Specifically, after performing the convolution operation, the threshold of the soft thresholding function is estimated using a soft thresholding module, and an adaptive slope module (ASB) is added to form a residual shrinking unit (RSBU-ASB) based on the adaptive slope module. The most suitable slope of the signal is automatically inferred using an attention mechanism to further correct the threshold.

[0038] When using the soft thresholding module for threshold estimation, a one-dimensional vector is first obtained by applying a gap to the absolute value of the feature map. This one-dimensional vector is then passed to two fully connected (FC) layers to obtain the scale parameter α. c Finally, the sigmoid function is used to scale the scale parameter to the range (0, 1). The formula for calculating the scale parameter is: Formula Six Among them, z c For the characteristics of neurons in layer c, τ c The threshold is obtained by multiplying the scale parameter by the absolute value of the one-dimensional feature vector, i.e. Formula 7; Where i, j, and c are the width, height, and channel of the feature map x, respectively.

[0039] A2. An attention mechanism is introduced into the network, and an adaptive slope module is added to automatically obtain the optimal threshold for denoising seismic signals, thereby achieving optimal processing of different signals and different levels of noise. In the adaptive slope module, the optimal slope for the signal is automatically inferred using an attention mechanism, and its output is: Formula 8; Where β is the adaptive slope factor, q c These are the characteristics of neurons in layer c; Finally, the output of the soft thresholding module is corrected based on the adaptive slope factor β to obtain the optimal soft threshold, i.e.: , Form Nine.

[0040] Based on the obtained optimal soft threshold, features with absolute values ​​less than the threshold are deleted, while features with absolute values ​​greater than the threshold are moved closer to 0, in order to achieve the purpose of shrinkage.

[0041] B. Establish an earthquake detection model to accurately obtain the detection results of earthquake events.

[0042] In step B, the established earthquake detection model includes a training set, a validation set, and a test set. The samples are randomly allocated to the established training set, validation set, and test set in a ratio of 8:1:1, respectively, for model training, model hyperparameter adjustment, and testing of the model's earthquake event detection performance.

[0043] In the training set, multiple iterations of training are set up. Through repeated training and optimization experiments, the error of the training results is minimized, the model converges, and the optimal model is obtained.

[0044] After the model training is completed, the established earthquake detection model is validated using samples from the test set.

[0045] Specifically, such as Figure 1As shown, this invention stacks a certain number of residual shrinking units (RSBU-ASB), convolutional layers, batch normalization (BN), activation functions (ReLU), global average pooling (GAP), and fully connected layers (FC) together to obtain a complete deep residual shrinking network. In this network, the first layer is the input layer, which takes a three-component seismic signal of length 6,000 (i.e., channel C=3, width W=6000, height H=1). Next is a convolutional layer with 12 channels, a kernel size of 3, and a stride of 1. Then, there are combinations of five residual shrinking units (RSBU-ASB) (with channel numbers of 12, 24, 48, 96, and 192), each combination containing two residual shrinking units (RSBU-ASB) with strides of 2 and 1, respectively. Finally, GAP is used to reduce the feature map to a 192×1 one-dimensional vector, and a fully connected layer is used to classify seismic events and noise signals. Deep residual shrinkage networks are based on residual networks. They insert a soft thresholding function as a nonlinear transformation layer into the deep network structure and apply a separate soft threshold to each channel of the feature map to remove noise-related features. This improves the model's ability to learn seismic event features from low signal-to-noise ratio seismic signals and ultimately achieves accurate seismic event detection under strong background noise.

[0046] When establishing the earthquake detection model, the first data source used was the Stanford Earthquake Dataset (STEAD), which consists of over one million data points (including earthquake events and noise signals) recorded by seismic stations worldwide from January 1984 to August 2018. Each earthquake waveform in this dataset is a 60-second, three-component signal with a sampling rate of 100Hz. This invention randomly selected 75,000 earthquake events and 75,000 noise signals to form the experimental dataset, and randomly divided it into training, validation, and test sets at a ratio of 80%, 10%, and 10%, respectively, for model training, model hyperparameter adjustment, and testing of the model's earthquake event detection performance. The earthquake events and noise signals in the experimental dataset are shown below. Figure 2 As shown.

[0047] like Figure 3 As shown, data source two uses microseismic monitoring data from a shale gas development area in southern Sichuan, China. This experiment addressed the seismic problems induced by hydraulic fracturing during shale gas extraction. Thirteen three-component geophones were deployed, and seismic records were collected from February 2017 to July 2018. Each seismic waveform in this dataset is a 60-second three-component signal with a sampling rate of 250Hz. Similarly, this invention also randomly selected 75,000 seismic events and 75,000 noise signals to form experimental datasets, and divided them into training, validation, and test sets at a ratio of 80%, 10%, and 10%, respectively. Figure 3It can be seen that, compared with natural earthquake signals, the signal-to-noise ratio of microseismic signals generated by hydraulic fracturing is significantly lower, and the weak microseismic events are almost submerged in strong background noise.

[0048] like Figure 4 As shown, the model was trained using the selected training data. Through repeated training and optimization experiments, the training error was minimized, the model converged, and the optimal model was obtained. The accuracy and loss function curves during the model training process are shown in the figures below. Figure 4 As shown, with increasing iterations, the model converges rapidly, the training and validation loss function values ​​gradually decrease, while the accuracy gradually increases. When the iteration reaches 80 times, the model's loss function value no longer decreases, and the accuracy no longer improves, indicating that the model is converging. Finally, the training and validation loss function values ​​stabilize at 0.390 and 0.392, respectively, with corresponding accuracies of 99.90% and 99.78%.

[0049] Then, the trained model was used to test 15,000 samples from data source one that were not involved in model training and validation, to obtain the detection performance of earthquake events. The results were compared with mainstream detection methods, as shown in the table below:

[0050] Table 1 Comparison of earthquake event detection results based on STEAD test data and such Figures 5-6 As shown, this invention, using a trained DRSN, clearly demonstrates that it correctly identified 14,862 samples, with 136 earthquakes misclassified as noise and 2 noise samples misclassified as earthquakes. The accuracy, precision, recall, and F1 score reached 99.08%, 99.97%, 98.19%, and 99.07%, respectively. These evaluation results clearly show that DRSN has a relatively optimal detection performance.

[0051] The trained model was then used to test microseismic event detection on 15,000 samples from data source two that were not involved in model training and validation. The results were compared with mainstream detection methods, and are shown in the table below:

[0052] Table 2 Comparison of detection results based on hydraulic fracturing microseismic monitoring data As shown in Table 2, similar to the trend of STEAD (Experimental Data 1), the detection performance of each deep learning method is better than that of the traditional STA / LTA. DRSN shows the best performance, correctly identifying 14,314 samples. 437 seismic events were misclassified as noise signals, and 249 noise signals were misclassified as seismic events. Its accuracy, precision, recall, and F1-score reached 95.43%, 96.59%, 94.17%, and 95.37%, respectively, demonstrating the best detection performance.

[0053] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A low signal-to-noise ratio seismic detection model based on a deep residual shrinkage network, characterized in that: Includes the following steps: A. Construct a deep residual shrinkage network by stacking a certain number of residual shrinkage units based on adaptive slope modules, input layer, convolutional layer, batch normalization, ReLU activation function, global average pooling and cross-entropy error function together to construct a complete deep residual shrinkage network; A1. A soft thresholding function is inserted as a nonlinear transformation layer into the deep residual shrinkage network structure to filter out the influence of noise-related features; A2. An attention mechanism is introduced into the network, and an adaptive slope module is added to automatically obtain the optimal threshold for denoising seismic signals, thereby achieving optimal processing of different signals and different levels of noise. B. Establish an earthquake detection model to accurately obtain the detection results of earthquake events; In step A, the convolutional layer extracts features from the input object and multiplies the output features of the previous layer with the convolutional kernel according to certain rules. Formula 1; in To output the j-th feature of the l-th layer in the feature map, For the i-th feature in layer l-1 of the input feature map, Let the kernel size be the convolution kernel size between the j-th feature of layer l and the i-th feature of layer l-1. For the corresponding bias, * indicates the convolution operation, M j This indicates that it is used to calculate the j-th feature; In step A, after extracting the features of the input object through the convolutional layer, the extracted data is then normalized through batch normalization to improve the convergence speed and generalization ability of the convolutional neural network. A stretching parameter γ and an offset parameter are also introduced. To recover the feature distribution learned by the network layers; in Formula 2; Formula 3; Formula 4; Formula 5; In the formula, m is the batch size of the sample, and x i and y i These are the i-th input feature and output feature in the sample, respectively; After performing the convolution operation, the threshold of the soft thresholding function is estimated using a soft thresholding module. An adaptive slope module is added to form a residual shrinking unit based on the adaptive slope module. The most suitable slope of the signal is automatically inferred using an attention mechanism to further correct the threshold. When using the soft thresholding module for threshold estimation, a one-dimensional vector is first obtained by applying a gap to the absolute value of the feature map. This one-dimensional vector is then passed to two fully connected (FC) layers to obtain the scale parameter α. c Finally, the sigmoid function is used to scale the scale parameter to between 0 and 1. The formula for calculating the scale parameter is: Formula 6; Among them, z c For the characteristics of neurons in layer c, τ c The threshold is obtained by multiplying the scale parameter by the average absolute value of the one-dimensional feature vector, i.e. Formula 7; Where m, n, and c are the width, height, and channels of the feature map x, respectively, and average represents the average value. In the adaptive slope module, the optimal slope for the signal is automatically inferred using an attention mechanism, and its output is: Formula 8; Where β is the adaptive slope factor, q c These are the characteristics of neurons in layer c; Finally, the output of the soft thresholding module is corrected based on the adaptive slope factor β to obtain the optimal soft threshold, i.e.: Form Nine; Based on the obtained optimal soft threshold, features with absolute values ​​less than the threshold are deleted, while features with absolute values ​​greater than the threshold are moved closer to 0, in order to achieve the purpose of shrinkage.

2. The low signal-to-noise ratio seismic detection model based on deep residual shrinkage network according to claim 1, characterized in that: In step A, a seismic signal of size C×W×1 is input through the input layer, where C is the number of channels, W is the width, and 1 is the height of the seismic signal. Then, a convolutional layer is used to extract the features of each channel of the signal. To prevent overfitting and increase the nonlinear expressive power of the model, batch normalization and ReLU activation function are added before the convolution operation.

3. The low signal-to-noise ratio seismic detection model based on a deep residual shrinkage network according to claim 1, characterized in that: The established earthquake detection model includes a training set, a validation set, and a test set. The samples are randomly allocated to the established training set, validation set, and test set in a ratio of 8:1:1, respectively, for model training, model hyperparameter adjustment, and testing of the model's earthquake event detection performance.

4. The low signal-to-noise ratio seismic detection model based on a deep residual shrinkage network according to claim 3, characterized in that: In the training set, multiple iterations of training are set up. Through repeated training and optimization experiments, the error of the training results is minimized, the model converges, and the optimal model is obtained. After the model is trained on the training set, the established earthquake detection model is verified on the samples in the test set.

Images