Earthquake first arrival wave pickup method and device based on multi-input attention mechanism network

Abstract

The invention discloses a seismic first arrival wave pickup method and device based on a multi-input attention mechanism network, multi-domain data and multiple network advantages are fused to realize first arrival automatic pickup, and compared with manual pickup, the efficiency is greatly improved, and the result is more objective; compared with a traditional automatic pickup method, the method is higher in precision and stability under different signal-to-noise ratios. Meanwhile, through hyper-parameter optimization, the generalization ability is high, reliable data can be provided for subsequent seismic processing, the advantages are outstanding under complex geological conditions, and the method is of great significance to geological analysis and resource exploration.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic data processing, and in particular to a method and device for picking seismic first arrival waves based on a multi-input attention mechanism network. Background Art

[0002] In the seismic data processing pipeline, first-arrival traveltime data is crucial for near-surface inversion and forms the cornerstone of subsequent analyses. Manual picking, a traditional method for obtaining first-arrival waves, while maintaining high accuracy, is not only time-consuming and labor-intensive, but also highly susceptible to the interpreter's experience. With the continued expansion of seismic exploration operations, the number of coverages has increased significantly, and the amount of seismic data has exploded, making manual picking increasingly inefficient and inaccurate.

[0003] Existing semi-automatic or automatic picking techniques leave much to be desired. Picking methods based on the cross-correlation characteristics of adjacent traces significantly affect the results of the selected standard traces, making it difficult to guarantee picking accuracy in seismic data processing scenarios with low signal-to-noise ratios. The long-short time window energy ratio (STA / LTA) method, with its energy-based algorithm, demonstrates good noise immunity and automatic picking performance. However, the window length setting significantly affects picking accuracy, leading to the development of various improved methods based on multiple time windows. The Akaike Information Criterion (AIC) method determines first arrivals by exploiting the statistical differences between seismic signals and noise, but often yields unsatisfactory picking results in low signal-to-noise ratio environments. While fractal dimension algorithms offer some noise immunity, they ignore the similarities between seismic traces. Furthermore, methods that transform the signal into other domains or spaces to highlight the difference between signal and noise, such as the τ-p transform and wavelet transform, while showing some advantages in picking first arrivals on different gathers, still struggle to achieve high-precision first arrival picking for complex seismic data.

[0004] Since the rise of deep learning, various network models have been introduced to the field of seismic first-break picking. However, current first-break picking efforts are generally limited to the time-space domain or a specific transform domain, failing to fully exploit the value of data features across multiple domains. Data features such as amplitude, phase, and frequency are key elements in identifying first-breaks. Failure to identify these features can lead to inaccurate and insufficient seismic first-break picking. Summary of the Invention

[0005] Based on the problems raised by the above background technology, the purpose of the present invention is to provide a method and device for picking earthquake first-arrival waves based on a multi-input attention mechanism network. The fusion features after fusion by the coordinate attention module are passed through two convolutional layers and activation functions in the output module to obtain the final first-arrival position prediction result, thereby overcoming the current situation that earthquake first-arrival picking is generally limited to a specific domain, while ignoring the characteristic value of seismic data in multiple domains.

[0006] The present invention is achieved through the following technical solutions: The first aspect of the present invention provides a method for picking earthquake first arrival waves based on a multi-input attention mechanism network, comprising the following steps: Step S1, acquiring seismic data, and preprocessing the seismic data to obtain preprocessed seismic data; the preprocessed seismic data includes time-space domain seismic data, instantaneous amplitude, frequency and phase characteristics; Step S2: construct a multi-input attention mechanism network architecture, wherein the multi-input attention mechanism network architecture includes four branch modules, a coordinate attention module and an output layer; Step S3, using the four branch modules to process the time-space domain seismic data, the instantaneous amplitude, the frequency, and the phase features respectively to obtain four multi-domain features; Step S4: Use the coordinate attention module to perform coordinate attention processing on the four multi-domain features to obtain fused features, and output the fused features through the output layer to obtain picking results.

[0007] In this technical solution, seismic data from different regions and geological conditions are first collected. This data covers a variety of complex geological structures and noisy environments. Because the collected seismic data is subject to noise interference and large amplitude variations, it is pre-processed to remove random noise and abnormal signals. At the same time, a normalization method is used to unify the amplitude of the seismic data to a specific range to improve data quality and provide a reliable data foundation for subsequent network training.

[0008] A multi-input attention mechanism network architecture is then constructed. Specifically, the MI-Attn-PrecisePickNet network is used to automatically pick the first arrivals of seismic signals. This network deeply mines the potential information of seismic data from multiple dimensions to improve the accuracy and reliability of first arrival picking. This multi-input attention network architecture includes four branch modules, which are used to process four types of two-dimensional input data: time-space domain seismic data, instantaneous amplitude, frequency, and phase features, respectively, to obtain four multi-domain features. This allows the time-space characteristics of seismic signals to be obtained from different perspectives, providing a foundation for subsequent feature extraction and analysis.

[0009] After being processed by the four branch modules, the four two-dimensional input data obtain features from different perspectives. In order to make full use of these multi-domain features, this method uses the coordinate attention module to apply different attention to them to achieve feature fusion. At the same time, it realizes the recalibration of features to highlight important features and suppress minor features.

[0010] The fused features after fusion by the coordinate attention module pass through two convolutional layers and activation functions in the output module to obtain the final first-arrival position prediction result, thereby overcoming the current problem that earthquake first-arrival picking is generally limited to a specific domain, while ignoring the characteristic value of earthquake data in multiple domains.

[0011] In an optional embodiment, the branch module includes: an m-layer residual module and an n-layer residual long short-term memory module; wherein the time-space domain seismic data, the instantaneous amplitude, the frequency and the phase characteristics are used as two-dimensional input data, and the two-dimensional input data are processed by the m-layer residual module to capture the spatial characteristics of the seismic signal; The spatial features of the seismic signal are converted into a time series, and the n-layer residual long short-term memory module is used to capture the features of the time series to obtain dynamic features in the time dimension.

[0012] In an optional embodiment, processing the two-dimensional input data using the m-layer residual module includes:

[0013] In the above formula, is the first The output of the layer, Representative The input data of the layer residual module, For the Nonlinear transformation of the layer residual module, For the Layer residual module parameters.

[0014] In an optional embodiment, using the n-layer residual long short-term memory module to capture features of the time series includes:

[0015] In the above formula, For the Layer The output hidden state after the residual connection at any moment, For the LSTM network, For the Layer residual long short-term memory module in The input sequence of moments.

[0016] In an optional embodiment, the coordinate attention module is used to perform coordinate attention processing on the four multi-domain features, including: Perform feature splicing on the four multi-domain features to obtain splicing features; Perform coordinate attention calculation on the spliced ​​features to obtain fusion features; wherein the formula for coordinate attention calculation is as follows:

[0017] In the above formula, and Usually the index of the feature map in height and width direction; and Represents the result of global pooling in horizontal and vertical directions, is the height, is the width; is the ReLU activation function, yes Convolutional layers; and It will The result after splitting is the feature map in horizontal and vertical directions; is the Sigmoid activation function, and It is used to generate horizontal and vertical attention weights Convolutional layers; Represents element-wise multiplication; The splicing feature.

[0018] In an optional embodiment, the training of the multi-input attention mechanism network architecture includes: Selecting samples from the earthquake data as a training set, and inputting the training set into the multi-input attention mechanism network architecture; The multi-input attention mechanism network architecture is trained using an adaptive matrix of the back-propagation method using binary cross entropy as the loss function; wherein the loss function is expressed as follows:

[0019] In the above formula, is the loss value, y is a binary label, which takes the value of 0 or 1; p(y) represents the probability that the output belongs to the y label; is the total number of training samples.

[0020] In an optional embodiment, after obtaining the picking result, the method further includes: Obtaining a test set label, and comparing the picking result with the test set label to obtain comparison data; The comparison data is marked with pixel precision to obtain the correct pixel ratio; wherein the calculation process of marking the comparison data with pixel precision is as follows:

[0021] In the above formula, For the correct pixel ratio, represents the number of pixels whose category i is inferred to be category i, Represents all pixels.

[0022] The second aspect of the present invention provides an earthquake first arrival wave picking system based on a multi-input attention mechanism network, comprising: A data processing module is used to acquire seismic data and preprocess the seismic data to obtain preprocessed seismic data; the preprocessed seismic data includes time-space domain seismic data, instantaneous amplitude, frequency and phase characteristics; An architecture building module for building a multi-input attention mechanism network architecture, wherein the multi-input attention mechanism network architecture includes four branch modules, a coordinate attention module, and an output layer; a branch feature module, configured to process the time-space domain seismic data, the instantaneous amplitude, the frequency, and the phase features respectively using the four branch modules to obtain four multi-domain features; A feature fusion module is used to use the coordinate attention module to perform coordinate attention processing on the four multi-domain features to obtain fused features, and output the fused features through the output layer.

[0023] The third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the computer program, an earthquake first arrival wave picking method based on a multi-input attention mechanism network is implemented.

[0024] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for picking earthquake first arrival waves based on a multi-input attention mechanism network.

[0025] Compared with the prior art, the present invention has the following advantages and beneficial effects: This method integrates multi-domain data and the advantages of various networks to achieve automatic first arrival picking, significantly improving efficiency and providing more objective results compared to manual picking. Compared to traditional automatic picking methods, it also achieves higher accuracy and stability across various signal-to-noise ratios. Through hyperparameter optimization, it has strong generalization capabilities and can provide reliable data for subsequent seismic processing. It is particularly advantageous in complex geological conditions and has significant implications for geological analysis and resource exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the examples. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be considered as limiting the scope. A person of ordinary skill in the art can also derive other relevant drawings based on these drawings without inventive effort. In the drawings: Figure 1 A schematic diagram of the structure of the multi-input attention mechanism network architecture provided in Example 1 of the present invention; Figure 2 Schematic diagram of the hyperparameter experiment provided in Example 1 of the present invention; wherein, Figure 2 (a) is a schematic diagram of the accuracy curve of different learning rates. Figure 2 (b) is a schematic diagram of the accuracy curve for different kernel matrix sizes. Figure 2 (c) is a schematic diagram of the accuracy curve for different batch sizes. Figure 2 (d) Schematic diagram of the accuracy curve for different model depths; Figure 3 This is a comparison chart of the first arrival picking results provided by Example 1 of the present invention; wherein, Figure 3 (a) is the original earthquake record map, Figure 3 (b) Comparison of picking results using different methods; Figure 4 This is an absolute error comparison chart provided in Example 1 of the present invention; Figure 5 Projections of the first arrival time on seismic data (a), instantaneous amplitude (b), frequency characteristic profile (c), and phase characteristic profile (d) provided in Example 1 of the present invention; Figure 6 This is a partial enlarged view of the seismic trace provided in Example 1 of the present invention; wherein, Figure 6 (a) is a partial enlarged image of the seismic trace of the original earthquake record. Figure 6 (b) is a partial magnified image of the seismic trace of the instantaneous amplitude. Figure 6 (c) is a partial magnified view of the seismic trace with frequency characteristics. Figure 6 (d) is a partial magnified view of the seismic trace with phase characteristics. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with examples and drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0028] Example 1 Embodiment 1 of the present invention provides a method for picking earthquake first arrival waves based on a multi-input attention mechanism network, the method comprising the following steps: Step S1, acquiring seismic data, and preprocessing the seismic data to obtain preprocessed seismic data; the preprocessed seismic data includes time-space domain seismic data, instantaneous amplitude, frequency and phase characteristics; Step S2: construct a multi-input attention mechanism network architecture, wherein the multi-input attention mechanism network architecture includes four branch modules, a coordinate attention module and an output layer; Step S3, using the four branch modules to process the time-space domain seismic data, the instantaneous amplitude, the frequency, and the phase features respectively to obtain four multi-domain features; Step S4: Use the coordinate attention module to perform coordinate attention processing on the four multi-domain features to obtain fused features, and output the fused features through the output layer to obtain picking results.

[0029] It's important to note that existing techniques for first-arrival picking are generally limited to the time-space domain or a specific transform domain, failing to fully exploit the value of data features across multiple domains. In fact, data features such as amplitude, phase, and frequency are key elements in identifying first arrivals. Failure to identify these features can lead to inaccurate and insufficient seismic wave first-arrival picking.

[0030] This example proposes a method for picking earthquake first-arrival waves based on a multi-input attention network. This method first collects seismic data from different regions and geological conditions, covering a variety of complex geological structures and noisy environments. Because the collected seismic data is subject to noise interference and large amplitude variations, it is pre-processed to remove random noise and abnormal signals. Furthermore, a normalization method is used to normalize the amplitude of the seismic data to a specific range, improving data quality and providing a reliable data foundation for subsequent network training.

[0031] A multi-input attention mechanism network architecture is then constructed. Specifically, the MI-Attn-PrecisePickNet network is used to automatically pick the first arrivals of seismic signals. This network deeply mines the potential information of seismic data from multiple dimensions to improve the accuracy and reliability of first arrival picking. This multi-input attention network architecture includes four branch modules, which are used to process four types of two-dimensional input data: time-space domain seismic data, instantaneous amplitude, frequency, and phase features, respectively, to obtain four multi-domain features. This allows the time-space characteristics of seismic signals to be obtained from different perspectives, providing a foundation for subsequent feature extraction and analysis.

[0032] After being processed by the four branch modules, the four two-dimensional input data obtain features from different perspectives. In order to make full use of these multi-domain features, this method uses the coordinate attention module to apply different attention to them to achieve feature fusion. At the same time, it realizes the recalibration of features to highlight important features and suppress minor features.

[0033] The fused features after fusion by the coordinate attention module pass through two convolutional layers and activation functions in the output module to obtain the final first-arrival position prediction result, thereby overcoming the current problem that earthquake first-arrival picking is generally limited to a specific domain, while ignoring the characteristic value of earthquake data in multiple domains.

[0034] Furthermore, the seismic data preprocessing also includes: using Hilbert-Huang transform to calculate the seismic data to obtain three-dimensional sections, which can reflect various characteristics of the seismic data, specifically: Analytical signal of seismic signal It can be expressed as:

[0035] in, is the real part of the complex channel, which is the actual earthquake record channel; is the imaginary part of the complex channel and is the orthogonal to the real part.

[0036] The instantaneous amplitude is a measure of the reflection strength and can be expressed as:

[0037] The instantaneous phase is a measure of the continuity of events on a seismic section:

[0038] The instantaneous frequency is the time rate of change of phase:

[0039] Significant changes in the three types of instantaneous information at the same location may reveal details that are not observable in the original time domain information. Since the first arrival is the first strong amplitude signal received by the detector in the seismic record, the amplitude is its primary characteristic. Seismic data may also contain noise with similar frequency and phase to the effective signal. To suppress the noise, the instantaneous frequency and instantaneous phase are constrained by amplitude (i.e., seismic data) to obtain phase characteristic data and frequency characteristic data:

[0040]

[0041] In an optional embodiment, the branch module includes: an m-layer residual module and an n-layer residual long short-term memory module; wherein the time-space domain seismic data, the instantaneous amplitude, the frequency and the phase characteristics are used as two-dimensional input data, and the two-dimensional input data are processed by the m-layer residual module to capture the spatial characteristics of the seismic signal; The spatial features of the seismic signal are converted into a time series, and the n-layer residual long short-term memory module is used to capture the features of the time series to obtain dynamic features in the time dimension.

[0042] It should be noted that the purpose of building the MI-Attn-PrecisePickNet network architecture is to automatically pick the first arrival of seismic signals and deeply explore the potential information of seismic data from multiple dimensions to improve the accuracy and reliability of first arrival picking. Figure 1 As shown in Figure 1, it includes a branch module, a coordinate attention module, and an output module. The branch module (dual residual feature learning module) consists of an m-layer residual module (ResNet) and an n-layer residual long short-term memory module (residual LSTM). The four branch modules process four types of two-dimensional input data: time-space domain seismic data, instantaneous amplitude, and frequency and phase features. This allows the temporal and spatial characteristics of seismic signals to be acquired from different perspectives, providing a foundation for subsequent feature extraction and analysis.

[0043] In this embodiment, weights are shared between different branch networks of the constructed MI-Attn-PrecisePickNet network architecture, which can effectively reduce the number of training parameters; multi-feature data is input into the network after Hilbert-Huang transform, and the attention module assigns different weights to amplitude, frequency, and phase features respectively; the network structure is flexible and can be applied to tasks of different complexity by adjusting the number of layers of feature extraction units.

[0044] Specifically, the residual module transfers the original two-dimensional seismic trace data directly to the output end through a shortcut path, completely preserving spatial information such as amplitude and phase; the convolution kernel slides to perceive local spatial relationships, maintain spatial order and position, and capture multi-scale features; at the same time, the network learns residual information, fuses the original spatial information with the convolution features, retains key details such as geological structure boundaries, and effectively preserves the spatial information of the seismic trace.

[0045] In an optional embodiment, processing the two-dimensional input data using the m-layer residual module includes:

[0046] In the above formula, is the first The output of the layer, Representative The input data of the layer residual module, For the Nonlinear transformation of the layer residual module, For the Layer residual module parameters.

[0047] Among them, The nonlinear transformation of the layer residual module includes convolution and activation operations.

[0048] After processing by m layers of residual modules, the output is converted into a sequence suitable for processing by the Residual Long Short-Term Memory (RSTM) module. The Residual Long Short-Term Memory (RSTM) module, with its unique gating mechanism (input gate, forget gate, and output gate), effectively processes time series. By selectively retaining or forgetting historical information, it accurately captures waveform, amplitude, and phase changes at different moments in the input data, thereby uncovering dynamic patterns in the data over time.

[0049] In an optional embodiment, using the n-layer residual long short-term memory module to capture features of the time series includes:

[0050] In the above formula, For the Layer The output hidden state after the residual connection at the moment, For the LSTM network, For the Layer residual long short-term memory module in The input sequence of moments.

[0051] Among them, Layer Output at any moment That is, the entire layer residual LSTM module is The output result at the moment, the final output result Converted to two-dimensional form , used for subsequent feature fusion and analysis.

[0052] The features after splicing are .

[0053] In an optional embodiment, the coordinate attention module is used to perform coordinate attention processing on the four multi-domain features, including: Perform feature splicing on the four multi-domain features to obtain splicing features; Perform coordinate attention calculation on the spliced ​​features to obtain fusion features; wherein the formula for coordinate attention calculation is as follows:

[0054] In the above formula, and Usually the index of the feature map in height and width direction; and Represents the result of global pooling in horizontal and vertical directions, is the height, is the width; is the ReLU activation function, yes Convolutional layers; and It will The result after splitting is the feature map in horizontal and vertical directions; is the Sigmoid activation function, and It is used to generate horizontal and vertical attention weights Convolutional layers; Represents element-wise multiplication; The splicing feature.

[0055] Feature map after fusion by coordinate attention module , and then through two convolutional layers and activation functions to obtain the final first arrival position prediction result:

[0056] In the above formula, and are the parameters of the two convolutional layers respectively.

[0057] Through the above formula, the network sequentially completes the feature extraction of input data, multi-domain feature capture, multi-domain feature fusion and final classification output. The output is consistent with the input size, meeting the requirements of seismic signal processing.

[0058] In an optional embodiment, the training of the multi-input attention mechanism network architecture includes: Selecting samples from the earthquake data as a training set, and inputting the training set into the multi-input attention mechanism network architecture; The multi-input attention mechanism network architecture is trained using an adaptive matrix of the back-propagation method using binary cross entropy as the loss function; wherein the loss function is expressed as follows:

[0059] In the above formula, is the loss value, y is a binary label, which takes the value of 0 or 1; p(y) represents the probability that the output belongs to the y label; is the total number of training samples.

[0060] When y is 0, it indicates a non-initial position, and when y is 1, it indicates an initial position.

[0061] Representative samples were selected from the collected seismic data, and professional seismic interpreters manually and accurately picked first arrivals to serve as real-world labeled data for network training. During the picking process, unified specifications were strictly adhered to to ensure accuracy and consistency. The time-space domain data, instantaneous amplitude characteristics, frequency and phase characteristics, and their corresponding first arrival picking information were organized into a training dataset and divided into a training set, a validation set, and a test set at an 80%, 10%, and 10% ratio. The training set was used to learn the parameters of the network model, the validation set was used to evaluate model performance and adjust hyperparameters during training, and the test set was used to ultimately evaluate the model's generalization ability and picking accuracy.

[0062] The training data is fed into the constructed MI-Attn-PrecisePickNet network, using supervised learning with binary cross-entropy as the loss function. Binary cross-entropy effectively measures the difference between the network's predictions and the true first-arrival labels. In the loss function calculation, for label y = 1, if the predicted value p(y) approaches 1, then the loss function should approach 0. Conversely, if the predicted value p(y) approaches 0, then the loss function should be very large.

[0063] The network is trained using the adaptive moment estimation (Adam) of the back-propagation method, with the goal of minimizing the loss function, which can complete the network model update. When the validation set loss decreases and remains stable within a certain iteration cycle, the training is stopped and the weights are adjusted. The result of the training run is saved. The saved optimal weights are used to predict the first arrivals on the test set, using the following formula:

[0064] Represents the role of the network.

[0065] During training, we closely monitor metrics such as validation set loss and accuracy. Based on these metrics, we dynamically adjust network training parameters, such as the learning rate, number of iterations, and batch size, to prevent overfitting and ensure good generalization. When validation set loss remains stable and no longer decreases within a certain number of iterations, we stop training and save the network weights at that point as the training result, which we will use for first-arrival prediction on the subsequent test set.

[0066] A systematic experimental analysis was conducted on the key hyperparameters of the MI-Attn-PrecisePickNet network, such as network depth, learning rate, convolution kernel size, batch size, etc. During the experiment, only one hyperparameter to be tested was changed each time, and the other parameters remained unchanged (e.g., the initial setting of learning rate was 0.005; convolution kernel size was 5; batch size was 4; network structure was ), by comparing the performance of the network under different hyperparameter combinations on the validation set, based on pixel accuracy, the hyperparameter model that can achieve the best performance of the network is determined. Among them, this embodiment experiments on the learning rate, kernel matrix size, batch size and network depth, and only changes one experimental parameter each time, while the other parameters remain constant, and the optimal parameters are determined based on the pixel accuracy of the validation set. The experimental results are shown in the figure. Figure 2 As shown, from Figure 2 (a) It can be seen that when the learning rate is too large, the network is difficult to converge and oscillation occurs. For this experimental data, a learning rate of 0.005 is the best. Figure 2 (b) shows that when the kernel matrix size is 5, the network accuracy is the highest and the convergence speed is fast; Figure 2 (c) It can be seen that although a larger batch size can improve generalization ability, it also increases computational time and memory requirements; Figure 2 (d) display, and All network structures can achieve good accuracy, but increasing network depth will increase memory usage. Through this experiment, we can effectively confirm the parameters used in model training.

[0067] The trained and hyperparameter-optimized MI-Attn-PrecisePickNet network model is used to process real earthquake data in both the time-space and time-frequency domains. The network output is a probability value indicating that each data point is a first arrival location. This output is thresholded, and traces with probabilities less than 0.4 are discarded from first arrival picking. For traces with probabilities greater than or equal to 0.4, the point with the highest probability is selected as the final first arrival location and marked as 1, thus enabling automatic first arrival picking.

[0068] This embodiment is aimed at the 3D seismic data collected in the North China Plain, such as Figure 3 As shown in (a), the 3D seismic data contains 250 shot set records, each shot set has 10 arrays, each array has 60 channels, each channel has 501 sampling points, the sampling rate is 4ms, and the surface elevation is between 92 and 160m. In order to reduce the memory consumption of training, the original data and the amplitude, frequency, and phase characteristic profiles are cut into 256×256 data blocks along the time and space directions. Adjacent data blocks partially overlap, the input batch size is set to 4, and the Adam algorithm is used for optimization. The data is normalized before entering the network and divided into training set, validation set, and test set at 80%, 10%, and 10%. The energy ratio method, manual picking, and MI-Attn-PrecisePickNet method are used to compare the results of first arrival picking. The comparison results are shown in the figure. Figure 3 (b) shows that in areas with high signal-to-noise ratios (SNRs), the MI-Attn-PrecisePickNet method and manual picking yield similar results. However, when the seismic record contains strong anomalous noise, MI-Attn-PrecisePickNet fails to pick first arrivals because the training labels do not include the first arrival positions of these seismic traces. In low SNR conditions, manual picking, which relies on inferences based on the results of surrounding traces, appears to be more effective, but it is subject to subjective factors. While the energy ratio method picks every trace, the results are disorganized at low SNRs. Given that false first arrivals significantly impact velocity modeling, while a small number of missing first arrivals have little impact on subsequent processing, it is more reasonable to abandon picking at low SNRs rather than accept false first arrivals.

[0069] Figure 4 The absolute error of the seismic traces picked by this method relative to the manual picking results is provided. The absolute error shows that the energy ratio method performs poorly compared to MI-Attn-PrecisePickNet.

[0070] The projection of the first arrival time on the seismic data, instantaneous amplitude, frequency characteristic section and phase characteristic section is as follows: Figure 5 shown.

[0071] like Figure 6 As shown in the local enlarged view of tracks 29-35, it can be seen that the first arrival is located at the first continuous strong amplitude position and has similar phase and frequency characteristics. The method proposed in this embodiment can accurately pick out the first arrival through these characteristics.

[0072] In an optional embodiment, after obtaining the picking result, the method further includes: Obtaining a test set label, and comparing the picking result with the test set label to obtain comparison data; The comparison data is marked with pixel precision to obtain the correct pixel ratio; wherein the calculation process of marking the comparison data with pixel precision is as follows:

[0073] In the above formula, For the correct pixel ratio, represents the number of pixels whose category i is inferred to be category i, Represents all pixels.

[0074] This metric is used to comprehensively evaluate the model's picking performance. Furthermore, tests are conducted on actual seismic data with varying signal-to-noise ratios to analyze the model's stability and accuracy under varying data quality conditions, fully verifying the model's reliability and applicability.

[0075] Based on the evaluation results, the network model is further verified and optimized. If the model is found to perform poorly under certain data characteristics or geological scenarios, such as low picking accuracy in low signal-to-noise ratio data, or unsatisfactory seismic data picking results for specific geological structures (such as near faults or in complex lithologic areas), targeted adjustments to the network structure can be made, such as increasing or decreasing the number of residual module layers, adjusting the number of hidden layer nodes in the residual long-short-term memory module, and optimizing the weight distribution method of the attention mechanism. Training data can be reselected to supplement data samples with specific geological characteristics to enhance the model's adaptability to various complex situations. Training parameters can also be optimized, such as adjusting the learning rate and batch size, to improve the model's convergence speed and accuracy. Through continuous verification and optimization, the model's first-arrival picking accuracy and stability will be continuously improved, enabling it to be better applied in actual seismic data processing and providing more reliable first-arrival data support for seismic exploration and geological research.

[0076] The method provided in this embodiment achieves the following effects: 1. Construct a multi-domain fusion earthquake first-arrival picking network: By integrating time-space domain and time-frequency domain data, a characteristic response relationship that can accurately identify earthquake first arrivals is established; 2. Achieve high-precision first-arrival picking: Utilize a multi-domain fusion network to efficiently and accurately pick first arrivals from massive seismic data, eliminating the picking errors of traditional methods in complex data environments; 3. Improve the overall efficiency of seismic processing: Provide reliable first-arrival data for subsequent seismic processing processes such as near-surface inversion and seismic imaging, optimize seismic data processing results, and promote the development of intelligent seismic exploration.

[0077] Example 2 Embodiment 2 of the present invention provides an earthquake first arrival wave picking system based on a multi-input attention mechanism network, the system comprising: A data processing module is used to acquire seismic data and preprocess the seismic data to obtain preprocessed seismic data; the preprocessed seismic data includes time-space domain seismic data, instantaneous amplitude, frequency and phase characteristics; An architecture building module for building a multi-input attention mechanism network architecture, wherein the multi-input attention mechanism network architecture includes four branch modules, a coordinate attention module, and an output layer; a branch feature module, configured to process the time-space domain seismic data, the instantaneous amplitude, the frequency, and the phase features respectively using the four branch modules to obtain four multi-domain features; A feature fusion module is used to use the coordinate attention module to perform coordinate attention processing on the four multi-domain features to obtain fused features, and output the fused features through the output layer.

[0078] Example 3

[0079] Embodiment 3 of the present invention provides an electronic device, which includes a processor, a memory, an input device, and an output device; the number of processors in the computer device can be one or more, taking one processor as an example; the processor, memory, input device, and output device in the electronic device can be connected via a bus or other means, taking connection via a bus as an example.

[0080] The memory, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules. The processor executes the software programs, instructions, and modules stored in the memory to perform various functional applications and data processing of the electronic device, thereby implementing the earthquake first-arrival wave picking method based on a multi-input attention mechanism network in Example 1.

[0081] The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data generated based on the use of the terminal, etc. Furthermore, the memory may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories may be connected to the electronic device via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The input device can be used to receive the ID and password input by the user, etc. The output device is used to output the network configuration page.

[0082] Example 4 Example 4 of the present invention also provides a computer-readable storage medium, and the computer-executable instructions, when executed by a computer processor, are used to implement the earthquake first arrival wave picking method based on a multi-input attention mechanism network as provided in Example 1.

[0083] An embodiment of the present invention provides a storage medium containing computer-executable instructions, and its computer-executable instructions are not limited to the method operations provided in Example 1, but can also execute related operations in the earthquake first arrival wave picking method based on a multi-input attention mechanism network provided in any embodiment of the present invention.

[0084] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for picking earthquake first arrival waves based on a multi-input attention mechanism network, characterized in that: The steps include: Step S1, acquiring seismic data, and preprocessing the seismic data to obtain preprocessed seismic data; the preprocessed seismic data includes time-space domain seismic data, instantaneous amplitude, frequency and phase characteristics; Step S2: construct a multi-input attention mechanism network architecture, wherein the multi-input attention mechanism network architecture includes four branch modules, a coordinate attention module and an output layer; Step S3, using the four branch modules to process the time-space domain seismic data, the instantaneous amplitude, the frequency, and the phase features respectively to obtain four multi-domain features; Step S4: Use the coordinate attention module to perform coordinate attention processing on the four multi-domain features to obtain fused features, and output the fused features through the output layer to obtain picking results.

2. The earthquake first arrival wave picking method based on the multi-input attention mechanism network according to claim 1 is characterized in that: The branch module includes: an m-layer residual module and an n-layer residual long short-term memory module; wherein the time-space domain seismic data, the instantaneous amplitude, the frequency and the phase characteristics are used as two-dimensional input data, and the two-dimensional input data are processed by the m-layer residual module to capture the spatial characteristics of the seismic signal; The spatial features of the seismic signal are converted into a time series, and the n-layer residual long short-term memory module is used to capture the features of the time series to obtain dynamic features in the time dimension.

3. The earthquake first arrival wave picking method based on the multi-input attention mechanism network according to claim 2 is characterized in that: Processing the two-dimensional input data using the m-layer residual module includes: In the above formula, is the first The output of the layer, Representative The input data of the layer residual module, For the Nonlinear transformation of the layer residual module, For the Layer residual module parameters.

4. The earthquake first arrival wave picking method based on the multi-input attention mechanism network according to claim 1 is characterized in that: Capturing features of the time series using the n-layer residual long short-term memory module includes: In the above formula, For the Layer The output hidden state after the residual connection at the moment, For the LSTM network, For the Layer residual long short-term memory module in The input sequence of moments.

5. The earthquake first arrival wave picking method based on the multi-input attention mechanism network according to claim 1 is characterized in that: The coordinate attention module is used to perform coordinate attention processing on the four multi-domain features, including: Perform feature splicing on the four multi-domain features to obtain splicing features; Perform coordinate attention calculation on the spliced ​​features to obtain fusion features; wherein the formula for coordinate attention calculation is as follows: In the above formula, and Usually the index of the feature map in height and width direction; and Represents the result of global pooling in horizontal and vertical directions, is the height, is the width; is the ReLU activation function, yes Convolutional layers; and It will The result after splitting is the feature map in horizontal and vertical directions; is the Sigmoid activation function, and It is used to generate horizontal and vertical attention weights Convolutional layers; Represents element-wise multiplication; The splicing feature.

6. The earthquake first arrival wave picking method based on a multi-input attention mechanism network according to claim 1, characterized in that: The training of the multi-input attention network architecture includes: Selecting samples from the earthquake data as a training set, and inputting the training set into the multi-input attention mechanism network architecture; The multi-input attention mechanism network architecture is trained using an adaptive matrix of the back-propagation method using binary cross entropy as the loss function; wherein the loss function is expressed as follows: In the above formula, is the loss value, y is a binary label, which takes the value of 0 or 1; p(y) represents the probability that the output belongs to the y label; is the total number of training samples.

7. The earthquake first arrival wave picking method based on a multi-input attention mechanism network according to claim 1 is characterized in that: After getting the picking result, it also includes: Obtaining a test set label, and comparing the picking result with the test set label to obtain comparison data; The comparison data is marked with pixel precision to obtain the correct pixel ratio; wherein the calculation process of marking the comparison data with pixel precision is as follows: In the above formula, For the correct pixel ratio, represents the number of pixels whose category i is inferred to be category i, Represents all pixels.

8. Earthquake first arrival wave picking system based on multi-input attention mechanism network, characterized by: include: A data processing module, configured to obtain seismic data and preprocess the seismic data to obtain preprocessed seismic data; The pre-processed seismic data includes time-space domain seismic data, instantaneous amplitude, frequency and phase characteristics; An architecture building module for building a multi-input attention mechanism network architecture, wherein the multi-input attention mechanism network architecture includes four branch modules, a coordinate attention module, and an output layer; a branch feature module, configured to process the time-space domain seismic data, the instantaneous amplitude, the frequency, and the phase features respectively using the four branch modules to obtain four multi-domain features; A feature fusion module is used to use the coordinate attention module to perform coordinate attention processing on the four multi-domain features to obtain fused features, and output the fused features through the output layer.

9. An electronic device, characterized in that: It includes a memory, a processor and a computer program stored in the memory and capable of running on the processor, and when the processor executes the computer program, it implements the earthquake first arrival wave picking method based on a multi-input attention mechanism network as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the earthquake first arrival wave picking method based on a multi-input attention mechanism network as described in any one of claims 1 to 7.

Images