A microseismic signal p-wave pickup method based on a dual-channel conditional attention mechanism

Abstract

This invention relates to the field of microseismic monitoring technology in mining engineering, specifically disclosing a P-wave pickup method for microseismic signals based on a dual-channel conditional attention mechanism. The method first collects microseismic events in the mine, performs multiple filtering processes on the raw waveform data, and manually marks the P-wave arrival times. Next, it uses a short-window and long-window energy ratio algorithm to detect the trigger point location of the waveform signal, and performs data augmentation and standardization on the relevant dataset. Then, it constructs a dual-channel conditional attention model, trains it using the standardized dataset, and obtains the P-wave pickup model. Finally, it inputs the waveform of the microseismic event to be processed into the model to obtain the P-wave arrival time pickup result. This invention integrates waveform features and trigger point location features, and combines a conditional attention mechanism to enhance the extraction capability of key features, effectively solving the problems of low accuracy and poor robustness in P-wave pickup under complex noise environments.

Description

Technical Field

[0001] This invention relates to the field of microseismic monitoring technology in mining engineering, and specifically discloses a method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism. Background Technology

[0002] In the mining process, microseismic signal processing plays a crucial role in microseismic location and rockburst early warning. Microseismic signals are usually generated by the fracturing of rock masses during mining, and contain key information on the internal damage evolution of the rock mass. Accurate acquisition of P-wave arrival time is the core foundation of microseismic location and rockburst early warning, because P-waves are longitudinal waves in seismic waves, and their arrival accuracy directly affects the location accuracy of microseismic events and the accuracy of rockburst early warning.

[0003] Currently, P-wave detection mainly relies on traditional methods and deep learning-based methods. Traditional methods include short-window / long-window energy ratio (STA / LTA) and autoregressive (AR) models, the core of which lies in identifying the arrival of P-waves through manually set thresholds. However, this method has significant limitations: first, it relies on manually set thresholds, which can easily lead to false positives in complex noisy environments; second, its ability to detect weak signals is limited because manually set thresholds often cannot adapt to changes in signal strength under different environments; and third, these methods do not fully utilize the temporal correlation of signals, failing to fully leverage the temporal characteristics of the signal to enhance detection accuracy.

[0004] In recent years, single-channel models based on deep learning, such as Convolutional Neural Networks (CNN) and Long Short-Term Memory Networks (LSTM), have been introduced into the field of microseismic signal processing. These methods demonstrate great potential by learning the temporal and spectral characteristics of the signal to automatically extract the arrival time of P waves. However, existing deep learning methods also have shortcomings: on the one hand, they only utilize the characteristics of the signal itself and do not fully combine prior information such as STA / LTA trigger points, resulting in incomplete utilization of the temporal correlation of the signal; on the other hand, they lack attention mechanisms for the characteristics of microseismic signals, making it difficult to focus on effective P-wave features in strong noise backgrounds.

[0005] In summary, existing microseismic signal processing technologies have significant shortcomings in mining environments, especially in the accurate acquisition of P-wave arrival times under complex noise conditions. There is an urgent need for a new microseismic signal P-wave acquisition method based on a dual-channel conditional attention mechanism to improve the accuracy and robustness of microseismic signal processing. Summary of the Invention

[0006] This invention addresses the problems in existing microseismic signal processing techniques, such as reliance on manually set thresholds leading to misjudgments, limited ability to detect weak signals, lack of prior information fusion, difficulty in effective feature focusing under strong noise backgrounds, and poor P-wave picking accuracy. It proposes a microseismic signal P-wave picking method based on a dual-channel conditional attention mechanism.

[0007] This invention provides a method for picking up P-waves in microseismic signals based on a dual-channel conditional attention mechanism, comprising the following steps:

[0008] S1. Collect microseismic events from the mine to obtain the raw waveform dataset;

[0009] S2. Perform multiple filtering processes on the original waveforms in the original waveform dataset to obtain a waveform dataset composed of waveforms; manually pick up the P-wave arrival times of the original waveforms in the original waveform dataset and mark the P-wave arrival times to obtain a P-wave arrival time marked dataset.

[0010] S3. Detect the trigger point position of each waveform signal in the waveform dataset based on the short-window and long-window energy ratio algorithm to obtain the trigger point position dataset;

[0011] S4. Perform data augmentation and standardization on the waveform dataset, trigger point location dataset, and P-wave arrival time marker dataset respectively to obtain a standardized waveform dataset, a standardized trigger point location dataset, and a standardized P-wave arrival time marker dataset;

[0012] S5. Construct a dual-channel conditional attention model, which includes an input layer, a dual-channel feature module, a feature fusion and conditional attention module, an output layer, and a training module;

[0013] S6. Input the standardized waveform dataset, standardized trigger point location dataset, and standardized P-wave arrival time marker dataset into the dual-channel conditional attention model, train the dual-channel conditional attention model, and obtain the P-wave picking model after training is completed;

[0014] S7. Input the waveform of the microseismic event to be processed into the P-wave picking model to obtain the picking result when the P-wave arrives.

[0015] According to some embodiments of this application, a method for picking up P-waves of microseismic signals based on a dual-channel conditional attention mechanism is provided. In step S1, the first 6000 sampling points are extracted from each microseismic event as the original waveform. The corresponding duration of the sampling points is 1 second, and the sampling rate is 6000Hz. The portion of the original waveform with a signal length of less than 6000 sampling points is zero-padded to obtain the original waveform dataset.

[0016] According to some embodiments of this application, a method for picking up P-waves of microseismic signals based on a dual-channel conditional attention mechanism, in step S2, the multiple filtering process includes bandpass filtering, notch filtering, and Savitzky-Golay filtering. The bandpass filtering uses a fourth-order Butterworth filter with a passband frequency range of 2Hz~200Hz, and is used to suppress low-frequency mechanical vibration noise and high-frequency electromagnetic interference in the original waveform dataset. The notch filtering uses an IIR notch filter with a center frequency of 50Hz and a quality factor of 35, and is used to eliminate residual power frequency interference after bandpass filtering. The Savitzky-Golay filtering uses a smoothing filter with a window length of 101 and a polynomial order of 4, and is used to reduce random noise while preserving high-frequency details of the signal after notch filtering.

[0017] According to some embodiments of this application, a method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism, in step S3, the energy ratio algorithm of the short window and the long window has a short window length of 72 sampling points, each sampling point has a corresponding duration of 12ms, and a sampling rate of 6000Hz; the long window length has 900 sampling points, each sampling point has a corresponding duration of 150ms, and a sampling rate of 6000Hz; the energy ratio threshold is 2. The preset triggering condition for the trigger point position is that the energy ratio of three consecutive sampling points exceeds the energy ratio threshold, and the point with the largest ratio among the three consecutive sampling points is located as the trigger point position corresponding to the current filter.

[0018] According to some embodiments of this application, a method for picking up P-waves of microseismic signals based on a dual-channel conditional attention mechanism, in step S4, the data augmentation includes performing time offset augmentation on the waveform dataset, the trigger point location dataset, and the P-wave arrival time marker dataset respectively, maintaining the relative relationship between the trigger point location and the P-wave arrival time marker of the same waveform, expanding the amount of data, and obtaining the waveform augmented dataset, the trigger point location augmented dataset, and the P-wave arrival time marker augmented dataset;

[0019] The standardization process includes performing Z-score standardization on the waveform enhancement dataset to obtain a standardized waveform dataset; performing Gaussian distribution heatmap transformation on the trigger point location enhancement dataset to obtain a standardized trigger point location dataset; and mapping the P-wave arrival time markers in the P-wave arrival time marker enhancement dataset to the [-1,1] interval to obtain a standardized P-wave arrival time marker dataset.

[0020] According to some embodiments of this application, a method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism is provided. In step S5, the dual-channel feature module includes a main channel and an auxiliary channel. The main channel includes a first main convolutional layer, a first main pooling layer, a second main convolutional layer, a second main pooling layer, and a main bidirectional long short-term memory layer connected in sequence. The auxiliary channel includes a first auxiliary convolutional layer, a first auxiliary pooling layer, a second auxiliary convolutional layer, a second auxiliary pooling layer, and an auxiliary bidirectional long short-term memory layer connected in sequence.

[0021] The feature fusion and conditional attention module includes a conditional attention layer, a bidirectional long short-term memory layer, and a fully connected layer connected in sequence.

[0022] According to some embodiments of this application, a method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism, in step S5, the input layer includes a first channel and a second channel, the first channel being connected to the main channel and the second channel being connected to the auxiliary channel.

[0023] According to some embodiments of this application, a method for picking up P-waves of microseismic signals based on a dual-channel conditional attention mechanism, in step S6, the input of the main channel is the standardized waveform dataset, the main channel is used to extract waveform features, and the output is waveform features; the input of the auxiliary channel is the standardized trigger point location dataset, the auxiliary channel is used to extract trigger point location features, and the output is trigger point location features; the output of the output layer is the predicted P-wave arrival time data.

[0024] According to some embodiments of this application, a method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism, in step S6, the conditional attention layer fuses the waveform features with the trigger point location features to obtain fused features. The conditional attention layer dynamically weights the hidden state of the fused features using a normalized attention weight matrix to obtain weighted fused features, as shown in formula (1):

[0025] (1)

[0026] in, Represents the query vector. , Represents the fused feature matrix. , Indicate waveform characteristics, Representing the trigger point location features, the fused feature matrix The size is T rows × D in List, This indicates a query vector weight matrix. The size is D inLine × D out List, Represents the key vector. , Represents the key vector weight matrix, and queries the key vector weight matrix. The size is D in Line × D out List, Represents a value vector. , Represents the value vector weight matrix. The size is D in Line × D out List, This represents the normalized attention weight matrix;

[0027] The normalized attention weight matrix To normalize the attention weight matrix, we obtain the normalized attention weight matrix. As shown in formula (2):

[0028] (2)

[0029] in, express Normalization process Represents the attention weight matrix;

[0030] The attention weight matrix To calculate conditional attention weights for the fused features, a spatially constrained mask is used to dynamically adjust feature interactions, resulting in an attention weight matrix. As shown in formula (3):

[0031] (3)

[0032] in, Indicates the row number. Used to adjust the scale of the dot product. This represents the adaptive adjustment factor. Represents a spatially constrained mask;

[0033] The spatial constraint mask As shown in formula (4):

[0034] (4)

[0035] in, This represents any sampling point on the waveform. Indicates the trigger point location. Indicates the radius of interest.

[0036] According to some embodiments of this application, a method for picking up P-waves of microseismic signals based on a dual-channel conditional attention mechanism is provided. In step S6, the input of the training module is the predicted P-wave arrival time data and the standardized P-wave arrival time marker dataset. The training module is used to evaluate the performance of the dual-channel conditional attention model. When the performance of the dual-channel conditional attention model meets the set accuracy, the P-wave picking model is output.

[0037] This invention proposes a P-wave pickup method for microseismic signals based on a dual-channel conditional attention mechanism, addressing the accuracy bottleneck of P-wave arrival time pickup in deep mine microseismic monitoring. It achieves a breakthrough in the deep integration of traditional detection methods and data features. This application constructs a heterogeneous dual-channel model driven by both algorithm detection and data. Through cross-modal fusion of the main and auxiliary channels, it solves the problem of insufficient utilization of prior knowledge in traditional deep learning methods. An adaptive attention mechanism guided by the energy ratio algorithm detection constraint of short and long time windows is proposed. By dynamically balancing the spatial constraint mask and the weight allocation driven by data features through adjustable parameters, this mechanism significantly improves the model's ability to focus on features in the neighborhood of the trigger point location, maintaining error stability within ±20 sampling points even under noise interference, showing significant optimization compared to ordinary attention mechanisms. By fusing waveform features and trigger point location features, combined with the conditional attention mechanism to enhance the extraction capability of key features, it solves the problems of low accuracy and poor robustness of existing methods in complex noise environments for P-wave pickup. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating a microseismic signal P-wave pickup method based on a dual-channel conditional attention mechanism according to an embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram of the conditional attention mechanism of the conditional attention layer in an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram showing the distribution characteristics of picking errors of different models in embodiments of the present invention when the P-wave arrives. Detailed Implementation

[0041] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0042] Example 1: This example provides a method for picking up P-waves of microseismic signals based on a dual-channel conditional attention mechanism, such as... Figure 1 As shown, it includes the following steps:

[0043] S1. Collect microseismic events in the mine to obtain the raw waveform dataset.

[0044] S2. Perform multiple filtering processes on the original waveforms in the original waveform dataset to obtain a waveform dataset composed of waveforms; manually pick up the P-wave arrival times of the original waveforms in the original waveform dataset and mark the P-wave arrival times to obtain a P-wave arrival time marked dataset.

[0045] S3. Based on the short-window and long-window energy ratio algorithm, detect the trigger point position of each waveform signal in the waveform dataset to obtain the trigger point position dataset.

[0046] S4. Perform data augmentation and standardization on the waveform dataset, trigger point location dataset, and P-wave arrival time marker dataset respectively to obtain the standardized waveform dataset, standardized trigger point location dataset, and standardized P-wave arrival time marker dataset.

[0047] S5. Construct a dual-channel conditional attention model, which includes an input layer, a dual-channel feature module, a feature fusion and conditional attention module, an output layer, and a training module.

[0048] S6. Input the standardized waveform dataset, the standardized trigger point location dataset, and the standardized P-wave arrival time marker dataset into the dual-channel conditional attention model, and train the dual-channel conditional attention model. After training, the P-wave picking model is obtained. The standardized waveform dataset and the standardized trigger point location dataset are both training datasets, and the standardized P-wave arrival time marker dataset is the training label set.

[0049] S7. Input the waveform of the microseismic event to be processed into the P-wave picking model to obtain the picking result when the P-wave arrives.

[0050] Example 2: This example provides a method for picking up P-waves of microseismic signals based on a dual-channel conditional attention mechanism, including the following steps:

[0051] S1. Collect microseismic events in the mine to obtain the raw waveform dataset.

[0052] Specifically, in step S1, the first 6000 sampling points from each microseismic event are extracted as the original waveform. The corresponding duration of each sampling point is 1 second, and the sampling rate is 6000Hz. The portion of the original waveform with a signal length less than 6000 sampling points is zero-padded to obtain the original waveform dataset. More specifically, in this embodiment, 1930 microseismic events were selected.

[0053] S2. Perform multiple filtering processes on the original waveforms in the original waveform dataset to obtain a waveform dataset composed of waveforms; manually pick up the P-wave arrival times of the original waveforms in the original waveform dataset and mark the P-wave arrival times to obtain a P-wave arrival time marked dataset.

[0054] In step S2, the multiple filtering process includes bandpass filtering, notch filtering, and Savitzky-Golay filtering. The bandpass filtering uses a fourth-order Butterworth filter with a passband frequency range of 2Hz to 200Hz. The bandpass filtering is used to suppress low-frequency mechanical vibration noise and high-frequency electromagnetic interference in the original waveform dataset, retain the effective frequency band characteristics of the micro-vibration signal, and improve the signal-to-noise ratio. The notch filtering uses an IIR notch filter with a center frequency of 50Hz and a quality factor of 35. The notch filtering is used to eliminate the power frequency interference remaining after bandpass filtering. By adjusting the Q value, it achieves high selectivity suppression of narrowband noise and reduces the interference of power grid noise on the P-wave initial characteristics. The Savitzky-Golay filtering uses a smoothing filter with a window length of 101 and a polynomial order of 4. The Savitzky-Golay filtering is used to reduce random noise while retaining the high-frequency details of the signal after notch filtering. The Savitzky-Golay filtering achieves adaptive smoothing of the signal through sliding polynomial fitting, avoiding the distortion of the P-wave leading-edge characteristics by traditional filtering.

[0055] S3. Based on the short-window and long-window energy ratio algorithm, detect the trigger point position of each waveform signal in the waveform dataset to obtain the trigger point position dataset.

[0056] In step S3, the energy ratio algorithm for short and long time windows first uses a sliding window to calculate the energy ratio by selecting a short and long time window. The short time window has a length of 72 sampling points, with each sampling point having a corresponding duration of 12ms and a sampling rate of 6000Hz. The long time window has a length of 900 sampling points, with each sampling point having a corresponding duration of 150ms and a sampling rate of 6000Hz. The energy ratio threshold is 2. Then, the trigger point position is determined. The preset trigger condition for the trigger point position is that the energy ratio exceeds the energy ratio threshold for three consecutive sampling points. The point with the largest ratio among the three consecutive sampling points is located as the trigger point position corresponding to the current filter.

[0057] S4. Perform data augmentation and standardization on the waveform dataset, trigger point location dataset, and P-wave arrival time marker dataset respectively to obtain the standardized waveform dataset, standardized trigger point location dataset, and standardized P-wave arrival time marker dataset.

[0058] In step S4, data augmentation includes performing time offset augmentation on the waveform dataset, trigger point location dataset, and P-wave arrival time marker dataset respectively, maintaining the relative relationship between the trigger point location and the P-wave arrival time marker for the same waveform, and expanding the number of data to obtain the waveform augmented dataset, trigger point location augmented dataset, and P-wave arrival time marker augmented dataset; more specifically, in the time offset augmentation, the offset range is ±10 to ±40 sampling points, corresponding to ±1.67ms to ±6.67ms at a sampling rate of 6000Hz, expanding the original 1930 microseismic event samples to 5705 training sets, with each training set being a "standardized waveform - standardized trigger point location - standardized P-wave arrival time marker" sample pair.

[0059] The standardization process includes Z-score standardization of the waveform enhancement dataset to obtain a standardized waveform dataset; Gaussian distribution heatmap transformation of the trigger point location enhancement dataset to obtain a standardized trigger point location dataset; and mapping the P-wave arrival time markers in the P-wave arrival time marker enhancement dataset to the [-1,1] interval to obtain a standardized P-wave arrival time marker dataset.

[0060] S5. Construct a dual-channel conditional attention model, which includes an input layer, a dual-channel feature module, a feature fusion and conditional attention module, an output layer, and a training module.

[0061] In step S5, the dual-channel feature module includes a main channel and an auxiliary channel. The main channel consists of a first main convolutional layer, a first main pooling layer, a second main convolutional layer, a second main pooling layer, and a main bidirectional long short-term memory layer connected in sequence. The auxiliary channel consists of a first auxiliary convolutional layer, a first auxiliary pooling layer, a second auxiliary convolutional layer, a second auxiliary pooling layer, and an auxiliary bidirectional long short-term memory layer connected in sequence. The main and auxiliary channels are symmetrically designed and are both 1D convolutional-LSTM processing chains. The feature fusion and conditional attention module includes a conditional attention layer, a bidirectional long short-term memory layer, and a fully connected layer connected in sequence. The input layer includes a first channel and a second channel. The first channel is connected to the main channel, and the second channel is connected to the auxiliary channel.

[0062] More specifically, the input layer dimension is (6000, 2). The first channel input is a standardized waveform dataset, and the second channel input is a standardized trigger point location dataset. The kernel size of the first main convolutional layer and the first auxiliary convolutional layer is 15, and the stride is 4, both used to capture long-term features. The kernel size of the second main convolutional layer and the second auxiliary convolutional layer is 7, and the stride is 2, both used to capture short-term features. The max pooling factor of the first main pooling layer and the first auxiliary pooling layer is 4, and the max pooling factor of the second main pooling layer and the second auxiliary pooling layer is 2, used to gradually reduce the feature dimension and retain key information. The number of units in the main bidirectional long short-term memory layer and the auxiliary bidirectional long short-term memory layer is 128, both used to capture the bidirectional temporal dependencies of the input data. The width of the conditional attention layer is 40, meaning that each time step focuses on the contextual information of the 20 time steps before and after, and the activation function is softmax to ensure weight normalization and focus on high-energy P-wave feature regions. After fusing the features of the two branches, the conditional attention layer introduces a sequence self-attention mechanism, using the trigger point location as the conditional weight.

[0063] S6. Input the standardized waveform dataset, standardized trigger point location dataset, and standardized P-wave arrival time marker dataset into the dual-channel conditional attention model to train the dual-channel conditional attention model. After training, the P-wave picking model is obtained. The standardized waveform dataset and standardized trigger point location dataset are both training datasets, and the standardized P-wave arrival time marker dataset is the training label set.

[0064] More specifically, before input, the standardized waveform dataset, standardized trigger point location dataset, and standardized P-wave arrival time marker dataset are divided into training set, validation set, and test set, respectively. The training set (3993 groups), validation set (856 groups), and test set (856 groups) are divided in a ratio of 7:1.5:1.5, and a hierarchical random partitioning strategy is adopted to ensure the consistency of data distribution.

[0065] In step S6, the input of the main channel is a standardized waveform dataset, the main channel is used to extract waveform features, and the output is waveform features; the input of the auxiliary channel is a standardized trigger point location dataset, the auxiliary channel is used to extract trigger point location features, and the output is trigger point location features; the output of the output layer is the predicted P-wave arrival time data.

[0066] In step S6, the conditional attention layer fuses the waveform features with the trigger point location features to obtain fused features. The trigger point location is used as a spatial constraint embedded in the attention calculation. The conditional attention layer dynamically weights the hidden state of the fused features through a normalized attention weight matrix, so that the model focuses on the energy mutation features in the neighborhood of the trigger point location, and obtains the weighted fused features, as shown in formula (1):

[0067] (1)

[0068] in, Represents the query vector. , Represents the fused feature matrix. , Indicate waveform characteristics, Representing the trigger point location features, the fused feature matrix The size is T rows × D in List, This indicates a query vector weight matrix. The size is D in Line × D out List, Represents the key vector. , Represents the key vector weight matrix, and queries the key vector weight matrix. The size is D in Line × D out List, Represents a value vector. , Represents the value vector weight matrix. The size is D in Line × D out List, This represents the normalized attention weight matrix;

[0069] Normalized attention weight matrix To normalize the attention weight matrix, we obtain the normalized attention weight matrix. As shown in formula (2):

[0070] (2)

[0071] in, express Normalization processing Represents the attention weight matrix;

[0072] Attention weight matrix To calculate conditional attention weights for the fused features, a spatially constrained mask is used to dynamically adjust feature interactions, resulting in an attention weight matrix. As shown in formula (3):

[0073] (3)

[0074] in, Indicates the row number. Used to adjust the scale of the dot product. This represents the adaptive adjustment factor. Represents a spatially constrained mask;

[0075] Spatial Constraint Mask A spatial constraint mask centered on the trigger point location ttrigger, with a radius of interest of δ sampling points. To suppress noise interference in areas far from the trigger point, as shown in formula (4):

[0076] (4)

[0077] in, This represents any sampling point on the waveform. Indicates the trigger point location. Indicates the radius of interest.

[0078] The principle diagram of the conditional attention mechanism of the conditional attention layer is as follows: Figure 2 As shown.

[0079] The training module takes as input the predicted P-wave arrival time data and the standardized P-wave arrival time marker dataset. The training module is used to evaluate the performance of the dual-channel conditional attention model. When the performance of the dual-channel conditional attention model meets the set accuracy, the P-wave picking model is output.

[0080] The design of the conditional attention layer achieves synergistic optimization of data-driven and traditional testing algorithms, preserving the neural network's ability to abstract complex features while suppressing interference from irrelevant regions through prior knowledge of trigger point locations.

[0081] In addition, a regularization method is used, specifically L2 regularization (kernel_regularizer='l2') to prevent overfitting.

[0082] In the training of the dual-channel conditional attention model, the MSE loss function and Adam optimizer are used, combined with early stopping (patience=25), model checkpoints, and learning rate decay (factor=0.5, patience=15). During the optimization process, the batch size is 256 to balance memory usage and gradient update stability; the number of iterations is 250 epochs, and the early stopping strategy prevents overfitting; the learning rate is scheduled as follows: the initial learning rate is 1e-3, and the learning rate is halved when the MAE on the validation set does not decrease for 15 consecutive epochs.

[0083] S7. Input the waveform of the microseismic event to be processed into the P-wave picking model to obtain the picking result when the P-wave arrives.

[0084] This embodiment compares the performance of different models in picking up P-wave arrival times on the same dataset in terms of mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). The specific results are shown in Table 1.

[0085] Table 1 Performance of P-wave arrival time pickup results for different models

[0086]

[0087] As can be seen, the method in this embodiment achieves the best MAE, RMSE, and MAPE, reducing them by 8.5%, 5.9%, and 7.6% respectively compared to the best-performing single-channel conditional attention model in traditional methods. The method also demonstrates an order of magnitude higher error than the traditional STA / LTA method, validating the overwhelming advantage of the method in this embodiment. Figure 3 As shown, the distribution characteristics of the P-wave arrival picking errors for five models are illustrated. The error distribution of the P-wave picking model is highly concentrated within ±18 sampling points (85.76%), significantly better than the single-channel conditional attention model (80.28%) and the single-channel ordinary attention model (78.06%). The STA / LTA method has the widest error distribution, with only 17% of the errors concentrated within ±18 sampling points. Specifically, the dual-channel model exhibits symmetrical concentration characteristics in P-wave arrival picking, with the error peak located near zero. The STA / LTA method shows a systematic delay in P-wave arrival picking (mainly distributed between 18 and 36 sampling points).

[0088] Accurate acquisition of P-wave arrival times in mine microseismic signals is a core technical challenge for rock mass fracture mechanism analysis and disaster early warning. Addressing the bottlenecks of low accuracy in traditional STA / LTA methods under low signal-to-noise ratio environments and insufficient feature coupling in existing deep learning methods, this embodiment proposes a microseismic signal P-wave acquisition method based on a dual-channel conditional attention mechanism. It innovatively constructs a dual-channel input architecture: the main channel focuses on waveform time-frequency feature extraction, while the auxiliary channel embeds the trigger point location of STA / LTA (short-window and long-window energy ratio algorithm detection). The denoised waveform features and trigger point location features are processed separately, and the conditional attention mechanism guided by STA / LTA constraints is combined to enhance features during critical time periods. This method has significant application potential in fields such as real-time monitoring of dynamic disasters in deep mines.

[0089] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A method for picking up P-waves in microseismic signals based on a dual-channel conditional attention mechanism, characterized in that, Includes the following steps: S1. Collect microseismic events from the mine to obtain the raw waveform dataset; S2. Perform multiple filtering processes on the original waveforms in the original waveform dataset to obtain a waveform dataset composed of waveforms; manually pick up the P-wave arrival times of the original waveforms in the original waveform dataset and mark the P-wave arrival times to obtain a P-wave arrival time marked dataset. S3. Detect the trigger point position of each waveform signal in the waveform dataset based on the short-window and long-window energy ratio algorithm to obtain the trigger point position dataset; S4. Perform data augmentation and standardization on the waveform dataset, trigger point location dataset, and P-wave arrival time marker dataset respectively to obtain a standardized waveform dataset, a standardized trigger point location dataset, and a standardized P-wave arrival time marker dataset; S5. Construct a dual-channel conditional attention model, which includes an input layer, a dual-channel feature module, a feature fusion and conditional attention module, an output layer, and a training module; S6. Input the standardized waveform dataset, standardized trigger point location dataset, and standardized P-wave arrival time marker dataset into the dual-channel conditional attention model, train the dual-channel conditional attention model, and obtain the P-wave picking model after training is completed; S7. Input the waveform of the microseismic event to be processed into the P-wave picking model to obtain the picking result when the P-wave arrives.

2. The method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 1, characterized in that, In step S1, the first 6000 sampling points are extracted from each micro-vibration event as the original waveform. The corresponding duration of the sampling points is 1 second, and the sampling rate is 6000Hz. The portion of the original waveform with a signal length of less than 6000 sampling points is zero-padded to obtain the original waveform dataset.

3. The method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 1, characterized in that, In step S2, the multiple filtering process includes bandpass filtering, notch filtering, and Savitzky-Golay filtering. The bandpass filtering uses a fourth-order Butterworth filter with a passband frequency range of 2Hz to 200Hz. The bandpass filtering is used to suppress low-frequency mechanical vibration noise and high-frequency electromagnetic interference in the original waveform dataset. The notch filtering uses an IIR notch filter with a center frequency of 50Hz and a quality factor of 35. The notch filtering is used to eliminate residual power frequency interference after bandpass filtering. The Savitzky-Golay filtering uses a smoothing filter with a window length of 101 and a polynomial order of 4. The Savitzky-Golay filtering is used to reduce random noise while preserving high-frequency details of the signal after notch filtering.

4. The method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 1, characterized in that, In the energy ratio algorithm for the short and long time windows in step S3, the short time window has a length of 72 sampling points, each sampling point has a corresponding duration of 12ms, and the sampling rate is 6000Hz. The long time window has a length of 900 sampling points, each sampling point has a corresponding duration of 150ms, and the sampling rate is 6000Hz. The energy ratio threshold is 2. The preset triggering condition for the trigger point position is that the energy ratio exceeds the energy ratio threshold for three consecutive sampling points. The point with the largest ratio among the three consecutive sampling points is located as the trigger point position corresponding to the current filter.

5. The method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 1, characterized in that, In step S4, the data augmentation includes performing time offset augmentation on the waveform dataset, trigger point location dataset, and P-wave arrival time marker dataset respectively, maintaining the relative relationship between the trigger point location and the P-wave arrival time marker for the same waveform, expanding the amount of data, and obtaining the waveform augmented dataset, the trigger point location augmented dataset, and the P-wave arrival time marker augmented dataset. The standardization process includes performing Z-score standardization on the waveform enhancement dataset to obtain a standardized waveform dataset; and performing Gaussian distribution heatmap transformation on the trigger point location enhancement dataset to obtain a standardized trigger point location dataset. The P-wave arrival time markers in the P-wave arrival time marker augmentation dataset are mapped to the [-1,1] interval to obtain the standardized P-wave arrival time marker dataset.

6. The method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 1, characterized in that, In step S5, the dual-channel feature module includes a main channel and an auxiliary channel. The main channel includes a first main convolutional layer, a first main pooling layer, a second main convolutional layer, a second main pooling layer, and a main bidirectional long short-term memory layer connected in sequence. The auxiliary channel includes a first auxiliary convolutional layer, a first auxiliary pooling layer, a second auxiliary convolutional layer, a second auxiliary pooling layer, and an auxiliary bidirectional long short-term memory layer connected in sequence. The feature fusion and conditional attention module includes a conditional attention layer, a bidirectional long short-term memory layer, and a fully connected layer connected in sequence.

7. The method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 6, characterized in that, In step S5, the input layer includes a first channel and a second channel, the first channel being connected to the main channel and the second channel being connected to the auxiliary channel.

8. The method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 6, characterized in that, In step S6, the input of the main channel is the standardized waveform dataset, the main channel is used to extract waveform features, and the output is waveform features; the input of the auxiliary channel is the standardized trigger point location dataset, the auxiliary channel is used to extract trigger point location features, and the output is trigger point location features; the output of the output layer is the predicted P-wave arrival time data.

9. A method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 6, characterized in that, In step S6, the conditional attention layer fuses the waveform features with the trigger point location features to obtain fused features. The conditional attention layer dynamically weights the hidden states of the fused features using a normalized attention weight matrix to obtain weighted fused features, as shown in formula (1): （1） in, Represents the query vector. , Represents the fused feature matrix. , Indicate waveform characteristics, Representing the trigger point location features, the fused feature matrix The size is T rows × D in List, This indicates a query vector weight matrix. The size is D in Line × D out List, Represents the key vector. , Represents the key vector weight matrix, and queries the key vector weight matrix. The size is D in Line × D out List, Represents a value vector. , Represents the value vector weight matrix. The size is D in Line × D out List, This represents the normalized attention weight matrix; The normalized attention weight matrix To normalize the attention weight matrix, we obtain the normalized attention weight matrix. As shown in formula (2): （2） in, express Normalization process Represents the attention weight matrix; The attention weight matrix To calculate conditional attention weights for the fused features, a spatially constrained mask is used to dynamically adjust feature interactions, resulting in an attention weight matrix. As shown in formula (3): （3） in, Indicates the row number. Used to adjust the scale of the dot product. This represents the adaptive adjustment factor. Represents a spatially constrained mask; The spatial constraint mask As shown in formula (4): （4） in, This represents any sampling point on the waveform. Indicates the trigger point location. Indicates the radius of interest.

10. A method for picking up P-wave microseismic signals based on a dual-channel conditional attention mechanism according to claim 8, characterized in that, In step S6, the input to the training module is the predicted P-wave arrival time data and the standardized P-wave arrival time marker dataset. The training module is used to evaluate the performance of the dual-channel conditional attention model. When the performance of the dual-channel conditional attention model meets the set accuracy, the P-wave picking model is output.

Images