A method and device for processing microseismic acquisition data

Abstract

The present invention provides a microseismic acquisition data processing method and device, which utilizes multiple non-overlapping partitioning windows to segment seismic waveform data, and performs separate feature extraction on the window waveform data obtained by segmenting the multiple non-overlapping partitioning windows. In addition to the global waveform features obtained by feature extraction of the entire seismic waveform data, effective local waveform features can be extracted from the waveform data of each window, thereby supplementing more waveform feature information; after determining the seismic phase classification result based on the local waveform features or global waveform features extracted by the corresponding model, the probability of the seismic waveform data belonging to P waves, S waves and noise is obtained by multiplying the probabilities of the seismic waveform data belonging to P waves, S waves and noise in the seismic phase classification results output by each seismic phase classification model, thereby integrating the classification results of waveform features of different scales, reducing the interference of noise in the seismic waveform data, and improving the robustness and noise resistance of the seismic phase detection method.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and device for processing microseismic acquisition data. Background Art

[0002] As the core of earthquake early warning systems, earthquake phase detection plays a crucial role in the entire seismic data processing process. Seismic phase detection typically relies on manual classification by experienced professionals. While highly accurate, this process is also time-consuming and subject to significant subjective influence. With the rapid increase in the amount of seismic data, manual phase detection is no longer sufficient. Consequently, some work has proposed automated phase detection algorithms based on feature calculations, such as STA / LTA (Short-Time Average / Long-Time Average). However, these automated phase detection algorithms often utilize shallow features of seismic motion records for phase detection, which are easily affected by noise. Consequently, they suffer from insufficient robustness and relatively low accuracy for phase detection in low-signal-to-noise ratio data, such as microseismic data. Summary of the Invention

[0003] The present invention provides a microseismic data processing method and device to solve the defects of the prior art in the low signal-to-noise ratio data such as microseismic data, such as the lack of robustness and relatively low accuracy of phase detection.

[0004] The present invention provides a microseismic acquisition data processing method, comprising:

[0005] The seismic waveform data to be processed is segmented based on the window ranges of a plurality of non-overlapping divided windows to obtain a plurality of window waveform data; wherein the plurality of non-overlapping divided windows include a divided window having the same window range as the seismic waveform data and a divided window having a window length smaller than the seismic waveform data;

[0006] Inputting the window waveform data obtained by segmentation based on any divided window into a trained seismic phase classification model corresponding to the any divided window to obtain a seismic phase classification result output by the corresponding seismic phase classification model; the seismic phase classification result includes the probability of the corresponding window waveform data belonging to P wave, S wave and noise determined by the corresponding seismic phase classification model; the seismic phase classification model corresponding to each divided window is constructed based on a convolutional neural network with the same structure but different convolution kernel sizes;

[0007] Multiplying the probabilities of belonging to P waves, S waves, and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window respectively to obtain the probabilities that the seismic waveform data belongs to P waves, S waves, and noise;

[0008] The seismic phase type of the seismic waveform data is determined based on the probability that the seismic waveform data belongs to P-wave, S-wave and noise.

[0009] According to a microseismic acquisition data processing method provided by the present invention, the multiple non-overlapping partition windows are determined based on the following steps:

[0010] Training a seismic phase classification model corresponding to a fully divided window based on a first training set to obtain a first-stage seismic phase classification model corresponding to the fully divided window; the fully divided window is a divided window with the same window range as the seismic waveform data; the first training set includes sample seismic waveform data and its seismic phase labels;

[0011] Determining a plurality of candidate windows based on the time window of the sample seismic waveform data;

[0012] After dividing each sample seismic waveform data in the first training set into multiple sample window data based on the current candidate window in order of window length from large to small, the waveform characteristics of the sample window data of each sample seismic waveform data are extracted based on the first-stage seismic phase classification model corresponding to the fully divided window, and the current candidate window is screened based on the difference in waveform characteristics of the sample window data corresponding to the current candidate window and with different seismic phase types to obtain a divided window with a window length smaller than the seismic waveform data.

[0013] According to a microseismic acquisition data processing method provided by the present invention, the current candidate window is screened based on the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types to obtain a partitioned window whose window length is smaller than the seismic waveform data, specifically comprising:

[0014] Clustering is performed based on waveform features of sample window data corresponding to the current candidate window of each sample seismic waveform data to obtain multiple clusters corresponding to the current candidate window;

[0015] Determine the difference in the number of sample window data of different seismic phase types in the same cluster corresponding to the current candidate window as the difference in waveform characteristics of the sample window data of different seismic phase types corresponding to the current candidate window;

[0016] If the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types is greater than a preset difference threshold, the current candidate window is determined to be a partition window.

[0017] According to a microseismic acquisition data processing method provided by the present invention, if the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types is greater than a preset difference threshold, the current candidate window is determined to be a partition window, and then the method further includes:

[0018] Delete candidate windows whose window ranges overlap with the window range of the current candidate window.

[0019] According to a microseismic acquisition data processing method provided by the present invention, the determining of multiple candidate windows based on the time window of the sample seismic waveform data specifically includes:

[0020] Dividing the time window of the sample seismic waveform data into two candidate windows;

[0021] Repeatedly divide the candidate window with the smallest current window length into two new candidate windows until the window length of the candidate window with the smallest current window length reaches a preset length threshold.

[0022] According to a microseismic data processing method provided by the present invention, the trained seismic phase classification model corresponding to each divided window is obtained by training based on the following steps:

[0023] Determine, based on the seismic phase classification model corresponding to the full divided window, a seismic phase classification model corresponding to the divided window having a window length smaller than that of the seismic waveform data;

[0024] Based on the sample seismic waveform data and their phase labels in the second training set, the first-stage seismic phase classification model corresponding to the full-division window and the seismic phase classification model corresponding to the division window whose window length is smaller than the seismic waveform data are jointly trained to obtain the trained seismic phase classification models corresponding to each division window.

[0025] According to a microseismic acquisition data processing method provided by the present invention, the determining, based on the seismic phase classification model corresponding to the full divided window, of the seismic phase classification model corresponding to the divided window having a window length smaller than that of the seismic waveform data specifically includes:

[0026] Copying the seismic phase classification model corresponding to the full partition window to obtain an initial seismic phase classification model corresponding to the partition window whose window length is smaller than that of the seismic waveform data;

[0027] Based on the window length being smaller than the ratio between the window length of any divided window of the seismic waveform data and the window length of the full divided window, the convolution kernel size in the initial seismic phase classification model corresponding to any divided window is reduced to obtain the seismic phase classification model corresponding to any divided window.

[0028] According to a microseismic acquisition data processing method provided by the present invention, the seismic phase classification model corresponding to any divided window includes four convolutional layers and two fully connected layers.

[0029] According to a microseismic acquisition data processing method provided by the present invention, the seismic waveform data to be processed is obtained by segmenting the original seismic waveform data acquired at a frequency of 100 Hz based on a 4-second time window.

[0030] The present invention also provides a microseismic data processing device, comprising:

[0031] a window data division unit, configured to divide the seismic waveform data to be processed based on the window ranges of a plurality of non-overlapping divided windows to obtain a plurality of window waveform data; wherein the plurality of non-overlapping divided windows include a divided window having the same window range as the seismic waveform data and a divided window having a window length smaller than the seismic waveform data;

[0032] A parallel seismic phase classification unit is configured to input window waveform data obtained by segmentation based on any partitioned window into a trained seismic phase classification model corresponding to the partitioned window, thereby obtaining a seismic phase classification result output by the corresponding seismic phase classification model; the seismic phase classification result includes the probability of the corresponding window waveform data belonging to P waves, S waves, and noise, as determined by the corresponding seismic phase classification model; the seismic phase classification model corresponding to each partitioned window is constructed based on a convolutional neural network with the same structure but different convolution kernel sizes;

[0033] A classification result fusion unit is used to multiply the probabilities of belonging to P waves, S waves and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window, so as to obtain the probabilities that the seismic waveform data belongs to P waves, S waves and noise;

[0034] A seismic phase type determination unit is used to determine the seismic phase type of the seismic waveform data based on the probability that the seismic waveform data belongs to P wave, S wave and noise.

[0035] The present invention also provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-described microseismic acquisition data processing methods is implemented.

[0036] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the microseismic acquisition data processing method described in any one of the above is implemented.

[0037] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the above-mentioned microseismic acquisition data processing methods.

[0038] The present invention provides a microseismic acquisition data processing method and device, which divides the seismic waveform data to be processed by using multiple non-overlapping partitioning windows to reduce the noise in the waveform data of each window, and then performs separate feature extraction on the window waveform data obtained by partitioning the multiple non-overlapping partitioning windows. In addition to the global waveform features obtained by feature extraction of the entire seismic waveform data, effective local waveform features can be extracted from the waveform data of each window, thereby supplementing more waveform feature information; after determining the seismic phase classification result based on the local waveform features or global waveform features extracted by the corresponding model, the probability of the seismic waveform data belonging to P wave, S wave and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each partitioning window is multiplied respectively, thereby obtaining the probability that the seismic waveform data belongs to P wave, S wave and noise, and integrating the classification results of waveform features of different scales, reducing the interference of noise in the seismic waveform data, improving the robustness and noise resistance of the seismic phase detection method, and improving the phase detection accuracy of microseismic data. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 It is a flow chart of a microseismic acquisition data processing method provided by the present invention;

[0041] Figure 2 is a schematic diagram of window waveform data provided by the present invention;

[0042] Figure 3 1 is a flow chart of a method for determining a partitioned window provided by the present invention;

[0043] Figure 4 It is a structural schematic diagram of a microseismic acquisition data processing device provided by the present invention;

[0044] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0045] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0046] Figure 1 This is a flow chart of a microseismic data processing method provided by the present invention, such as Figure 1 As shown, the method includes:

[0047] Step 110, segmenting the seismic waveform data to be processed based on the window ranges of a plurality of non-overlapping divided windows to obtain a plurality of window waveform data; wherein the plurality of non-overlapping divided windows include a divided window having the same window range as the seismic waveform data and a divided window having a window length smaller than the seismic waveform data;

[0048] Step 120: Input the window waveform data obtained by segmentation based on any partitioned window into a trained seismic phase classification model corresponding to the partitioned window to obtain a seismic phase classification result output by the corresponding seismic phase classification model; the seismic phase classification result includes the probability of the corresponding window waveform data belonging to P waves, S waves, and noise, as determined by the corresponding seismic phase classification model; the seismic phase classification model corresponding to each partitioned window is constructed based on a convolutional neural network with the same structure but different convolution kernel sizes;

[0049] Step 130, multiplying the probabilities of belonging to P waves, S waves, and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window, respectively, to obtain the probabilities of the seismic waveform data belonging to P waves, S waves, and noise;

[0050] Step 140: Determine the phase type of the seismic waveform data based on the probability that the seismic waveform data belongs to P-wave, S-wave, and noise.

[0051] Specifically, features extracted by automated phase detection algorithms based on feature calculation, such as STA / LTA, are easily affected by noise. This leads to insufficient robustness and relatively low accuracy when performing phase detection on low signal-to-noise ratio data, such as microseismic data. To address this, embodiments of the present invention utilize the autonomous extraction of effective features by deep learning-based neural network models to improve their ability to resist noise interference when dealing with low signal-to-noise ratio data, such as microseismic data.

[0052] Among them, compared with the existing part of the work that directly uses the neural network model to extract the features of the seismic waveform data to be processed (such as the waveform data of the 4s to 60s time window) as a whole and performs seismic phase classification based on the extracted features, the embodiment of the present invention uses multiple non-overlapping partitioning windows (that is, there is no overlap between the window ranges of each partitioning window) to divide the seismic waveform data to be processed into multiple window waveform data, and performs separate feature extraction on the multiple window waveform data respectively. The reason for this is that there is a lot of noise in the seismic waveform data with a low signal-to-noise ratio. When the seismic waveform data is subjected to feature extraction as a whole, it is inevitable that it will be interfered by the noise, resulting in the extracted waveform features containing the features of the noise data, causing the subsequent classification results to be inaccurate. Therefore, the embodiment of the present invention uses multiple non-overlapping partitioning windows to divide the seismic waveform data to be processed, reduces the noise in each window waveform data, and then performs separate feature extraction on the window waveform data obtained by partitioning the multiple non-overlapping partitioning windows. In addition to the global waveform features obtained by feature extraction of the entire seismic waveform data, effective local waveform features can be extracted from each window waveform data, thereby supplementing more waveform feature information, so that the seismic phase detection method has higher robustness and anti-interference ability.

[0053] Specifically, the seismic waveform data to be processed can be three-component seismic wave data obtained by segmenting the original seismic waveform data collected at a frequency of 100 Hz based on a 4-second time window. The position of the window waveform data obtained by segmenting the seismic waveform data to be processed based on the window range of the partitioning window in the seismic waveform data corresponds to the window range of the corresponding partitioning window. Among them, a plurality of non-overlapping partitioning windows include partitioning windows with the same window range as the seismic waveform data and partitioning windows with a window length less than the seismic waveform data. That is, the partitioned window waveform data contains the seismic waveform data itself and several local data with a length shorter than the seismic waveform data. Taking the window ranges of each partitioning window as [0,2s], [2s,3s], [3s,4s] and [0,4s] as an example, the window waveform data obtained by segmentation are as follows: Figure 2 As shown in window waveform data 1, window waveform data 2, window waveform data 3 and window waveform data 4.

[0054] Subsequently, the window waveform data obtained by segmenting multiple non-overlapping partitioned windows are subjected to separate feature extraction and phase classification. Each partitioned window corresponds to its own phase classification model, which is used to receive the window waveform data obtained based on the corresponding partitioned window and output the phase classification result. Here, the phase classification result output by any phase classification model includes the probability that the input window waveform data determined by the phase classification model belongs to P wave (Primary wave), S wave (Secondary wave) and noise. Noise refers to types other than P wave and S wave.

[0055] Here, the seismic phase classification model corresponding to each divided window is constructed based on a convolutional neural network with the same structure and different convolution kernel sizes. In some embodiments, the seismic phase classification model corresponding to any divided window includes four convolution layers and two fully connected layers. Among them, the four convolution layers are used to extract features of the input window waveform data to obtain waveform features, and the two fully connected layers are used to perform classification based on the waveform features to obtain seismic phase classification results. The structures of the seismic phase classification models corresponding to different divided windows are the same, but the convolution kernel sizes in the four convolution layers are different. The larger the window length of the divided window (that is, the length of the window waveform data), the larger the convolution kernel size. It should be noted that setting the convolution kernel size of the seismic phase classification model to be adapted to the length of the window waveform data input therein can provide a receptive field adapted to the window waveform data, thereby extracting more complete waveform features.

[0056] The probabilities of belonging to P waves, S waves, and noise in the phase classification results output by the phase classification model corresponding to each partition window are multiplied respectively to obtain the probabilities that the seismic waveform data belongs to P waves, S waves, and noise. The following formula can be used to determine the probability P1 of the seismic waveform data belonging to P waves, the probability P2 of the seismic waveform data belonging to S waves, and the probability P3 of the seismic waveform data belonging to noise:

[0057]

[0058]

[0059]

[0060] in, and are the probabilities of belonging to P wave, S wave and noise in the phase classification results output by the phase classification model corresponding to the i-th partition window.

[0061] The seismic phase type of the seismic waveform data is determined based on the probabilities of the seismic waveform data belonging to P waves, S waves, and noise. If the probability that the seismic waveform data belongs to P waves is the highest among the three probabilities, the seismic waveform data is determined to belong to P waves (i.e., P phase); if the probability that the seismic waveform data belongs to S waves is the highest among the three probabilities, the seismic waveform data is determined to belong to S waves (i.e., S phase).

[0062] Here, the seismic phase classification results output by the seismic phase classification model corresponding to each divided window are determined based on the local waveform features or global waveform features extracted by the corresponding model. By multiplying the probabilities of belonging to P waves, S waves and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window, the probabilities of the seismic waveform data belonging to P waves, S waves and noise are obtained. The classification results of waveform features of different scales are integrated, the interference of noise in the seismic waveform data is reduced, the robustness and noise resistance of the seismic phase detection method are improved, and the phase detection accuracy of microseismic data is improved.

[0063] It can be seen that the method provided by the embodiment of the present invention divides the seismic waveform data to be processed by using multiple non-overlapping partitioning windows to reduce the noise in each window waveform data, and then performs separate feature extraction on the window waveform data obtained by partitioning multiple non-overlapping partitioning windows. In addition to the global waveform features obtained by feature extraction of the entire seismic waveform data, effective local waveform features can be extracted from each window waveform data, thereby supplementing more waveform feature information; after determining the seismic phase classification result based on the local waveform features or global waveform features extracted by the corresponding model, the probability of the seismic waveform data belonging to P wave, S wave and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each partitioning window is multiplied respectively, and the probability of the seismic waveform data belonging to P wave, S wave and noise is obtained, which integrates the classification results of waveform features of different scales, reduces the interference of noise in the seismic waveform data, improves the robustness and noise resistance of the seismic phase detection method, and improves the phase detection accuracy of microseismic data.

[0064] In order to ensure the effectiveness of the local waveform features extracted from the waveform data of each window, thereby improving the robustness and anti-noise ability of the phase detection method, the determination of the divided windows is the key. Figure 3 As shown, the multiple non-overlapping partition windows are determined based on the following steps:

[0065] Step 310: Training a seismic phase classification model corresponding to a fully divided window based on a first training set to obtain a first-stage seismic phase classification model corresponding to the fully divided window; the fully divided window is a divided window having the same window range as the seismic waveform data; the first training set includes sample seismic waveform data and its seismic phase labels;

[0066] Step 320 , determining a plurality of candidate windows based on the time window of the sample seismic waveform data;

[0067] Step 330, after dividing each sample seismic waveform data in the first training set into multiple sample window data based on the current candidate window in order of window length from large to small, extract the waveform characteristics of the sample window data of each sample seismic waveform data based on the first stage seismic phase classification model corresponding to the full division window, and screen the current candidate window based on the difference in waveform characteristics of the sample window data corresponding to the current candidate window and with different seismic phase types to obtain a division window with a window length smaller than the seismic waveform data.

[0068] Specifically, some training samples can be randomly divided from the acquired sample set to construct a first training set. Wherein, one training sample includes a sample seismic waveform data and its phase label. The sample seismic waveform data is obtained in the same way as the seismic waveform data above, and is obtained by segmenting the original seismic waveform data collected at a frequency of 100 Hz based on a 4-second time window. Based on the sample seismic waveform data and its phase label in the first training set, a supervised method is used to train the phase classification model corresponding to the full partition window to obtain the first-stage phase classification model corresponding to the full partition window. Wherein, the full partition window is a partition window with the same window range as the seismic waveform data. Here, some training samples can be randomly divided from the acquired sample set to construct a test set. During the above-mentioned training process of the phase classification model corresponding to the full partition window, the test set is used to evaluate the model. Specifically, the model can be used to classify the sample seismic waveform data in the test set to obtain the phase classification results of the corresponding sample seismic waveform data, and the accuracy of the above phase classification results can be statistically calculated based on the phase labels of each sample seismic waveform data. When the accuracy reaches a preset accuracy threshold (e.g., 80%), the training is temporarily stopped to obtain the first-stage seismic phase classification model corresponding to the full partition window. At this point, the first-stage seismic phase classification model corresponding to the full partition window has a certain feature extraction capability.

[0069] Based on the time window of the sample seismic waveform data, a plurality of candidate windows can be determined.

[0070] In some embodiments, the time window of the sample seismic waveform data can be divided into two equal parts, and then the candidate window with the smallest current window length (there may be 2, 4, or more) can be repeatedly divided into two new candidate windows until the window length of the candidate window with the smallest current window length reaches a preset length threshold. Taking the time window of the sample seismic waveform data as 4s and the preset length threshold as 1s as an example, the candidate windows [0, 2s], [2s, 4s], [0, 1s], [1s, 2s], [2s, 3s], and [3s, 4s] can be obtained in sequence.

[0071] Subsequently, each candidate window is processed in order of window length from largest to smallest. Specifically, each sample seismic waveform data in the first training set is segmented into multiple sample window data based on the current candidate window. A sample window data can be segmented from each sample seismic waveform data based on the current candidate window. The waveform features of the sample window data of each sample seismic waveform data are extracted based on the first-stage phase classification model corresponding to the fully segmented window. That is, after the sample window data of each sample seismic waveform data is input into the first-stage phase classification model corresponding to the fully segmented window, the output result of the last convolutional layer of the model is obtained as the waveform features of the sample window data of each sample seismic waveform data. The difference in waveform features of sample window data corresponding to the current candidate window and having different phase types (the phase type of the sample window data is the phase type of the sample seismic waveform data to which it belongs, which can be determined based on its phase label) is statistically analyzed. The greater the difference in waveform features of sample window data corresponding to the current candidate window and having different phase types, the more distinguishable the sample window data obtained by segmenting the candidate window is, the higher the possibility of extracting effective features from the sample window data or window waveform data obtained by segmenting the candidate window is, and the greater the help in improving the accuracy of phase classification. Therefore, the current candidate windows can be screened based on the above difference, and the candidate windows with larger difference are retained as the partition windows, thereby obtaining the partition windows with a window length smaller than the seismic waveform data.

[0072] In some embodiments, the step of screening the current candidate window based on the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types to obtain a partitioned window having a window length smaller than the seismic waveform data specifically includes:

[0073] Clustering is performed based on waveform features of sample window data corresponding to the current candidate window of each sample seismic waveform data to obtain multiple clusters corresponding to the current candidate window;

[0074] Determine the difference in the number of sample window data of different seismic phase types in the same cluster corresponding to the current candidate window as the difference in waveform characteristics of the sample window data of different seismic phase types corresponding to the current candidate window;

[0075] If the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types is greater than a preset difference threshold, the current candidate window is determined to be a partition window.

[0076] Specifically, the waveform characteristics of the sample window data of the current candidate window corresponding to each sample seismic waveform data can be clustered based on a clustering algorithm (such as DBSCAN, Meanshift, etc.) to obtain a plurality of clusters corresponding to the current candidate window. Each cluster contains sample window data corresponding to the current candidate window that are classified together by the clustering algorithm. Subsequently, the difference in the number of sample window data of different phase types in the same cluster corresponding to the current candidate window is determined as the difference in the waveform characteristics of the sample window data of different phase types corresponding to the current candidate window. Here, for any cluster corresponding to the current candidate window, the number of sample window data of different phase types in the cluster can be determined based on the phase labels of each sample window data corresponding to the current candidate window (that is, the phase labels of the sample seismic waveform data to which it belongs), and the difference in the number of sample window data of different phase types in the cluster is calculated. Specifically, the difference between the number of P-wave type sample window data and the number of S-wave type sample window data in the cluster, the difference between the number of P-wave type sample window data and the number of noise type sample window data in the cluster, and the difference between the number of S-wave type sample window data and the number of noise type sample window data in the cluster can be calculated, and the three differences can be added together to obtain the difference in the number of sample window data of different seismic phase types in the cluster. Then, based on the difference in the number of sample window data of different seismic phase types in each cluster corresponding to the current candidate window, the maximum value is determined, and the maximum value is used as the difference in waveform characteristics of the sample window data of different seismic phase types corresponding to the current candidate window.

[0077] If the difference in waveform characteristics of the sample window data corresponding to the current candidate window and having different seismic phase types is greater than a preset difference threshold, the current candidate window is determined to be a split window. In addition, to ensure that there is no overlap between the window ranges of the screened split windows and to improve the efficiency of split window screening, after the current candidate window is determined to be a split window, candidate windows that have not yet been processed and whose window ranges overlap with the window range of the current candidate window can be deleted.

[0078] In other embodiments, the trained seismic phase classification model corresponding to each divided window is obtained by training based on the following steps:

[0079] Determine, based on the seismic phase classification model corresponding to the full divided window, a seismic phase classification model corresponding to the divided window having a window length smaller than that of the seismic waveform data;

[0080] Based on the sample seismic waveform data and their phase labels in the second training set, the first-stage seismic phase classification model corresponding to the full-division window and the seismic phase classification model corresponding to the division window whose window length is smaller than the seismic waveform data are jointly trained to obtain the trained seismic phase classification models corresponding to each division window.

[0081] Specifically, since the seismic phase classification models corresponding to different partitioning windows have the same model structure but different convolution kernel sizes, the seismic phase classification model corresponding to the full partitioning window can be directly reused to obtain an initial seismic phase classification model corresponding to the partitioning window with a window length less than the seismic waveform data, and then the convolution kernel size of the initial seismic phase classification model corresponding to the partitioning window with a window length less than the seismic waveform data is adjusted to obtain a seismic phase classification model corresponding to the partitioning window with a window length less than the seismic waveform data. When adjusting the convolution kernel size of the initial seismic phase classification model corresponding to the partitioning window with a window length less than the seismic waveform data, the convolution kernel size in the initial seismic phase classification model corresponding to the partitioning window can be reduced based on the ratio between the window length of the corresponding partitioning window and the window length of the full partitioning window to obtain the seismic phase classification model corresponding to the partitioning window. For example, if the ratio between the window length of the corresponding partitioning window and the window length of the full partitioning window is 1 / 2, the convolution kernel size in the initial seismic phase classification model corresponding to the partitioning window is proportionally reduced, and the reduced convolution kernel size is 1 / 2 of the convolution kernel size of the seismic phase classification model corresponding to the full partitioning window (if the size value after proportional reduction is a decimal, it is rounded).

[0082] Based on the sample seismic waveform data and its phase labels in the second training set randomly divided from the sample set, the phase classification model of the first stage corresponding to the full partition window and the phase classification model corresponding to the partition window with a window length less than the seismic waveform data are jointly trained to obtain the trained phase classification model corresponding to each partition window. Specifically, the same sample seismic waveform data can be input into the phase classification model of the first stage corresponding to the full partition window and the phase classification model corresponding to the partition window with a window length less than the seismic waveform data to obtain the phase classification results of the sample seismic waveform data output by each model. Subsequently, using the method given in the above embodiment, the probability of belonging to P wave, S wave and noise in the phase classification results of the sample seismic waveform data output by each model is multiplied respectively to obtain the probability that the sample seismic waveform data belongs to P wave, S wave and noise, and then the cross entropy is solved based on the probability of each sample seismic waveform data belonging to P wave, S wave and noise and the phase label of each sample seismic waveform data to obtain the model loss, and the parameters of each model are adjusted based on the model loss to achieve joint training of multiple models.

[0083] A microseismic acquisition data processing device provided by the present invention is described below. The microseismic acquisition data processing device described below and the microseismic acquisition data processing method described above can be referenced to each other.

[0084] Based on any of the above embodiments, Figure 4 This is a structural diagram of a microseismic data acquisition and processing device provided by the present invention. Figure 4 As shown, the device includes: a window data division unit 410, a parallel seismic phase classification unit 420, a classification result fusion unit 430 and a seismic phase type determination unit 440.

[0085] The window data division unit 410 is configured to divide the seismic waveform data to be processed based on the window ranges of a plurality of non-overlapping divided windows to obtain a plurality of window waveform data; wherein the plurality of non-overlapping divided windows include divided windows having the same window range as the seismic waveform data and divided windows having a window length less than that of the seismic waveform data;

[0086] The parallel phase classification unit 420 is used to input the window waveform data obtained based on the segmentation of any divided window into the trained phase classification model corresponding to the any divided window, and obtain the phase classification result output by the corresponding phase classification model; the phase classification result includes the probability of the corresponding window waveform data belonging to P wave, S wave and noise determined by the corresponding phase classification model; the phase classification model corresponding to each divided window is constructed based on a convolutional neural network with the same structure but different convolution kernel sizes;

[0087] The classification result fusion unit 430 is used to multiply the probabilities of belonging to P waves, S waves and noise in the seismic phase classification results output by the seismic phase classification models corresponding to each divided window, and obtain the probabilities that the seismic waveform data belongs to P waves, S waves and noise;

[0088] The seismic phase type determination unit 440 is used to determine the seismic phase type of the seismic waveform data based on the probability that the seismic waveform data belongs to P wave, S wave and noise.

[0089] The device provided by the embodiment of the present invention divides the seismic waveform data to be processed by using multiple non-overlapping partitioning windows to reduce the noise in each window waveform data, and then performs separate feature extraction on the window waveform data obtained by partitioning the multiple non-overlapping partitioning windows. In addition to the global waveform features obtained by feature extraction of the entire seismic waveform data, effective local waveform features can be extracted from each window waveform data, thereby supplementing more waveform feature information; after determining the seismic phase classification result based on the local waveform features or global waveform features extracted by the corresponding model, the probability of the seismic waveform data belonging to P wave, S wave and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each partitioning window is multiplied respectively, thereby obtaining the probability that the seismic waveform data belongs to P wave, S wave and noise, and integrating the classification results of waveform features of different scales, reducing the interference of noise in the seismic waveform data, improving the robustness and noise resistance of the seismic phase detection method, and improving the phase detection accuracy of microseismic data.

[0090] Based on any of the above embodiments, the multiple non-overlapping partitioned windows are determined based on the following steps:

[0091] Training a seismic phase classification model corresponding to a fully divided window based on a first training set to obtain a first-stage seismic phase classification model corresponding to the fully divided window; the fully divided window is a divided window with the same window range as the seismic waveform data; the first training set includes sample seismic waveform data and its seismic phase labels;

[0092] Determining a plurality of candidate windows based on the time window of the sample seismic waveform data;

[0093] After dividing each sample seismic waveform data in the first training set into multiple sample window data based on the current candidate window in order of window length from large to small, the waveform characteristics of the sample window data of each sample seismic waveform data are extracted based on the first-stage seismic phase classification model corresponding to the fully divided window, and the current candidate window is screened based on the difference in waveform characteristics of the sample window data corresponding to the current candidate window and with different seismic phase types to obtain a divided window with a window length smaller than the seismic waveform data.

[0094] Based on any of the above embodiments, the current candidate window is screened based on the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types to obtain a partitioned window having a window length smaller than the seismic waveform data, specifically comprising:

[0095] Clustering is performed based on waveform features of sample window data corresponding to the current candidate window of each sample seismic waveform data to obtain multiple clusters corresponding to the current candidate window;

[0096] Determine the difference in the number of sample window data of different seismic phase types in the same cluster corresponding to the current candidate window as the difference in waveform characteristics of the sample window data of different seismic phase types corresponding to the current candidate window;

[0097] If the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types is greater than a preset difference threshold, the current candidate window is determined to be a partition window.

[0098] Based on any of the above embodiments, if the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types is greater than a preset difference threshold, then determining the current candidate window as a split window further includes:

[0099] Delete candidate windows whose window ranges overlap with the window range of the current candidate window.

[0100] Based on any of the above embodiments, determining multiple candidate windows based on the time window of the sample seismic waveform data specifically includes:

[0101] Dividing the time window of the sample seismic waveform data into two candidate windows;

[0102] Repeatedly divide the candidate window with the smallest current window length into two new candidate windows until the window length of the candidate window with the smallest current window length reaches a preset length threshold.

[0103] Based on any of the above embodiments, the trained seismic phase classification models corresponding to each divided window are obtained by training based on the following steps:

[0104] Determine, based on the seismic phase classification model corresponding to the full divided window, a seismic phase classification model corresponding to the divided window having a window length smaller than that of the seismic waveform data;

[0105] Based on the sample seismic waveform data and their phase labels in the second training set, the first-stage seismic phase classification model corresponding to the full-division window and the seismic phase classification model corresponding to the division window whose window length is smaller than the seismic waveform data are jointly trained to obtain the trained seismic phase classification models corresponding to each division window.

[0106] Based on any of the above embodiments, determining the seismic phase classification model corresponding to the divided window having a window length smaller than that of the seismic waveform data based on the seismic phase classification model corresponding to the full divided window specifically includes:

[0107] Copying the seismic phase classification model corresponding to the full partition window to obtain an initial seismic phase classification model corresponding to the partition window whose window length is smaller than that of the seismic waveform data;

[0108] Based on the window length being smaller than the ratio between the window length of any divided window of the seismic waveform data and the window length of the full divided window, the convolution kernel size in the initial seismic phase classification model corresponding to any divided window is reduced to obtain the seismic phase classification model corresponding to any divided window.

[0109] Based on any of the above embodiments, the seismic phase classification model corresponding to any divided window includes four convolutional layers and two fully connected layers.

[0110] Based on any of the above embodiments, the seismic waveform data to be processed is obtained by segmenting the original seismic waveform data collected at a frequency of 100 Hz based on a 4-second time window.

[0111] Figure 5 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 5As shown, the electronic device may include: a processor (processor) 510, a memory (memory) 520, a communication interface (Communications Interface) 530 and a communication bus 540, wherein the processor 510, the memory 520, and the communication interface 530 communicate with each other through the communication bus 540. The processor 510 can call the logic instructions in the memory 520 to execute a microseismic acquisition data processing method, which includes: segmenting the seismic waveform data to be processed based on the window range of multiple non-overlapping divided windows to obtain multiple window waveform data; wherein the multiple non-overlapping divided windows include divided windows with the same window range as the seismic waveform data and divided windows with a window length less than the seismic waveform data; inputting the window waveform data obtained by segmentation based on any divided window into a trained seismic phase classification model corresponding to any divided window to obtain a seismic phase classification result output by the corresponding seismic phase classification model; the seismic phase classification result includes the probability of the corresponding window waveform data belonging to P wave, S wave and noise determined by the corresponding seismic phase classification model; the seismic phase classification model corresponding to each divided window is constructed based on a convolutional neural network with the same structure and different convolution kernel sizes; multiplying the probabilities of belonging to P wave, S wave and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window respectively to obtain the probability of the seismic waveform data belonging to P wave, S wave and noise; and determining the seismic phase type of the seismic waveform data based on the probability of the seismic waveform data belonging to P wave, S wave and noise.

[0112] In addition, the logic instructions in the above-mentioned memory 520 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0113] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute a microseismic acquisition data processing method provided by the above methods, the method comprising: segmenting the seismic waveform data to be processed based on the window range of multiple non-overlapping divided windows to obtain multiple window waveform data; wherein the multiple non-overlapping divided windows include divided windows with the same window range as the seismic waveform data and divided windows with a window length smaller than the seismic waveform data; the window waveform data obtained based on the segmentation of any divided window is segmented; The shape data is input into the trained seismic phase classification model corresponding to any divided window to obtain the seismic phase classification result output by the corresponding seismic phase classification model; the seismic phase classification result includes the probability that the waveform data of the corresponding window belongs to P wave, S wave and noise determined by the corresponding seismic phase classification model; the seismic phase classification model corresponding to each divided window is constructed based on a convolutional neural network with the same structure and different convolution kernel sizes; the probabilities of belonging to P wave, S wave and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window are multiplied respectively to obtain the probability that the seismic waveform data belongs to P wave, S wave and noise; based on the probability that the seismic waveform data belongs to P wave, S wave and noise, the seismic phase type of the seismic waveform data is determined.

[0114] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the above-mentioned microseismic acquisition data processing method, the method comprising: segmenting the seismic waveform data to be processed based on the window range of multiple non-overlapping divided windows to obtain multiple window waveform data; wherein the multiple non-overlapping divided windows include divided windows with the same window range as the seismic waveform data and divided windows with a window length smaller than the seismic waveform data; inputting the window waveform data obtained by segmentation based on any divided window into the trained waveform data corresponding to any divided window The seismic phase classification model is used to obtain the seismic phase classification result output by the corresponding seismic phase classification model; the seismic phase classification result includes the probability that the waveform data of the corresponding window belongs to P wave, S wave and noise determined by the corresponding seismic phase classification model; the seismic phase classification model corresponding to each divided window is constructed based on a convolutional neural network with the same structure and different convolution kernel sizes; the probabilities of belonging to P wave, S wave and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window are multiplied respectively to obtain the probability that the seismic waveform data belongs to P wave, S wave and noise; based on the probability that the seismic waveform data belongs to P wave, S wave and noise, the seismic phase type of the seismic waveform data is determined.

[0115] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0116] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A microseismic data processing method, characterized in that: include: The seismic waveform data is segmented based on a plurality of non-overlapping segmented windows and a segmented window having the same window range as the seismic waveform data to be processed, to obtain a plurality of window waveform data; wherein the plurality of non-overlapping segmented windows include a segmented window having a window length smaller than that of the seismic waveform data; Inputting the window waveform data obtained by segmentation based on any divided window into a trained seismic phase classification model corresponding to the any divided window to obtain a seismic phase classification result output by the corresponding seismic phase classification model; the seismic phase classification result includes the probability of the corresponding window waveform data belonging to P wave, S wave and noise determined by the corresponding seismic phase classification model; the seismic phase classification model corresponding to each divided window is constructed based on a convolutional neural network with the same structure but different convolution kernel sizes; Multiplying the probabilities of belonging to P waves, S waves, and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window respectively to obtain the probabilities that the seismic waveform data belongs to P waves, S waves, and noise; determining a seismic phase type of the seismic waveform data based on probabilities that the seismic waveform data belongs to a P wave, an S wave, and noise; The partitioning window is determined based on the following steps: Training a seismic phase classification model corresponding to a fully divided window based on a first training set to obtain a first-stage seismic phase classification model corresponding to the fully divided window; the fully divided window is a divided window with the same window range as the seismic waveform data; the first training set includes sample seismic waveform data and its seismic phase labels; Determining a plurality of candidate windows based on the time window of the sample seismic waveform data; After dividing each sample seismic waveform data in the first training set into multiple sample window data based on the current candidate window in order of window length from large to small, the waveform characteristics of the sample window data of each sample seismic waveform data are extracted based on the first-stage seismic phase classification model corresponding to the fully divided window, and the current candidate window is screened based on the difference in waveform characteristics of the sample window data corresponding to the current candidate window and with different seismic phase types to obtain a divided window with a window length smaller than the seismic waveform data.

2. A microseismic data processing method according to claim 1, characterized in that: The method of screening the current candidate window based on the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types to obtain a partitioned window having a window length smaller than the seismic waveform data specifically includes: Clustering is performed based on waveform features of sample window data corresponding to the current candidate window of each sample seismic waveform data to obtain multiple clusters corresponding to the current candidate window; Determine the difference in the number of sample window data of different seismic phase types in the same cluster corresponding to the current candidate window as the difference in waveform characteristics of the sample window data of different seismic phase types corresponding to the current candidate window; If the difference in waveform characteristics of sample window data corresponding to the current candidate window and having different seismic phase types is greater than a preset difference threshold, the current candidate window is determined to be a partition window.

3. The microseismic data processing method according to claim 2, characterized in that: If the difference in waveform features of the sample window data corresponding to the current candidate window and having different seismic phase types is greater than a preset difference threshold, the current candidate window is determined to be a split window, and then the following further comprises: Delete candidate windows whose window ranges overlap with the window range of the current candidate window.

4. The microseismic data processing method according to claim 1, characterized in that: The determining of a plurality of candidate windows based on the time window of the sample seismic waveform data specifically includes: Dividing the time window of the sample seismic waveform data into two candidate windows; Repeatedly divide the candidate window with the smallest current window length into two new candidate windows until the window length of the candidate window with the smallest current window length reaches a preset length threshold.

5. The microseismic data processing method according to claim 1, characterized in that: The trained seismic phase classification models corresponding to each partition window are obtained based on the following training steps: Determine, based on the seismic phase classification model corresponding to the full divided window, a seismic phase classification model corresponding to the divided window having a window length smaller than that of the seismic waveform data; Based on the sample seismic waveform data and their phase labels in the second training set, the first-stage seismic phase classification model corresponding to the full-division window and the seismic phase classification model corresponding to the division window whose window length is smaller than the seismic waveform data are jointly trained to obtain the trained seismic phase classification models corresponding to each division window.

6. A microseismic data processing method according to claim 5, characterized in that: The determining, based on the seismic phase classification model corresponding to the full divided window, the seismic phase classification model corresponding to the divided window with a window length smaller than that of the seismic waveform data specifically includes: Copying the seismic phase classification model corresponding to the full partition window to obtain an initial seismic phase classification model corresponding to the partition window whose window length is smaller than that of the seismic waveform data; Based on the window length being smaller than the ratio between the window length of any divided window of the seismic waveform data and the window length of the full divided window, the convolution kernel size in the initial seismic phase classification model corresponding to any divided window is reduced to obtain the seismic phase classification model corresponding to any divided window.

7. The microseismic data processing method according to claim 1, characterized in that: The seismic phase classification model corresponding to any divided window includes four convolutional layers and two fully connected layers.

8. The microseismic data processing method according to claim 1, characterized in that: The seismic waveform data to be processed is obtained by segmenting the original seismic waveform data collected at a frequency of 100 Hz based on a 4-second time window.

9. A microseismic data processing device, characterized in that: include: a window data division unit, configured to divide the seismic waveform data based on a plurality of non-overlapping division windows and a window range of a division window having the same window range as the seismic waveform data to be processed, to obtain a plurality of window waveform data; wherein the plurality of non-overlapping division windows includes a division window having a window length smaller than that of the seismic waveform data; A parallel seismic phase classification unit is configured to input window waveform data obtained by segmentation based on any partitioned window into a trained seismic phase classification model corresponding to the partitioned window, thereby obtaining a seismic phase classification result output by the corresponding seismic phase classification model; the seismic phase classification result includes the probability of the corresponding window waveform data belonging to P waves, S waves, and noise, as determined by the corresponding seismic phase classification model; the seismic phase classification model corresponding to each partitioned window is constructed based on a convolutional neural network with the same structure but different convolution kernel sizes; A classification result fusion unit is used to multiply the probabilities of belonging to P waves, S waves and noise in the seismic phase classification results output by the seismic phase classification model corresponding to each divided window, so as to obtain the probabilities that the seismic waveform data belongs to P waves, S waves and noise; a seismic phase type determination unit, configured to determine the seismic phase type of the seismic waveform data based on the probability that the seismic waveform data belongs to a P wave, an S wave, and noise; The partitioning window is determined based on the following steps: Training a seismic phase classification model corresponding to a fully divided window based on a first training set to obtain a first-stage seismic phase classification model corresponding to the fully divided window; the fully divided window is a divided window with the same window range as the seismic waveform data; the first training set includes sample seismic waveform data and its seismic phase labels; Determining a plurality of candidate windows based on the time window of the sample seismic waveform data; After dividing each sample seismic waveform data in the first training set into multiple sample window data based on the current candidate window in order of window length from large to small, the waveform characteristics of the sample window data of each sample seismic waveform data are extracted based on the first-stage seismic phase classification model corresponding to the fully divided window, and the current candidate window is screened based on the difference in waveform characteristics of the sample window data corresponding to the current candidate window and with different seismic phase types to obtain a divided window with a window length smaller than the seismic waveform data.

Images