Microseismic event identification and classification method, device, equipment and storage medium

Abstract

The present invention discloses a method for identifying and classifying microseismic events, including extracting characteristic data of effective signals from microseismic signals to be analyzed; and identifying microseismic events on the characteristic data through a pre-created neural network recognition model to determine the microseismic events corresponding to the effective signals; and processing the effective signal samples according to at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending to increase the number of samples of effective signal samples; thereby obtaining enough samples for neural network training and obtaining a neural network recognition model. In the present application, bandpass filtering, adding Gaussian noise, window slicing, window bending, etc. are used to increase the number of samples, ensure the accuracy of the neural network recognition model, and then the accuracy of the neural network recognition model in identifying microseismic events. The present application also provides a microseismic event recognition and classification device, equipment, and computer-readable storage medium, which have the above-mentioned beneficial effects.

Description

Technical Field

[0001] The present invention relates to the technical field of microseismic event monitoring, and in particular to a method, device, equipment and computer-readable storage medium for identifying and classifying microseismic events. Background Art

[0002] The safety of coal mines, oil wells and other sites has always been an issue that needs to be focused on during the mining of coal mines, oil, and natural gas. Usually, microseismic events are caused by coal seam fractures, stratum structural activation, groundwater movement, and other reasons in the strata of coal mines and well sites. Through real-time monitoring of microseismic events, the specific cause of the microseismic event, that is, the specific type of microseismic time, is determined, which is an important reference for warning the safety of the site.

[0003] However, how to accurately and quickly identify microseismic events is one of the key issues currently being addressed in the industry. Summary of the invention

[0004] The purpose of the present invention is to provide a method, device, equipment and computer-readable storage medium for identifying and classifying microseismic events, so as to improve the efficiency and accuracy of microseismic event identification to a certain extent.

[0005] In order to solve the above technical problems, the present invention provides a method for identifying and classifying microseismic events, comprising:

[0006] Extract characteristic data of effective signals from microseismic signals to be analyzed;

[0007] Identify microseismic events on the characteristic data using a pre-created neural network recognition model to determine the microseismic events corresponding to the effective signal;

[0008] The process of pre-creating the neural network recognition model includes:

[0009] Obtain valid signal samples with microseismic event labels;

[0010] Processing the valid signal samples according to at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending, and using the valid signal samples before and after the processing as valid signal samples to increase the number of valid signal samples;

[0011] Performing feature extraction on the effective signal sample to obtain a feature data sample;

[0012] A neural network training is performed based on the characteristic data samples and the microseismic event labels carried therein to obtain the neural network recognition model.

[0013] Optionally, performing neural network training according to the characteristic data sample and the earthquake event label carried therein includes:

[0014] An LSTM network is created, and the characteristic data samples and the earthquake event labels are input to perform neural network training.

[0015] Optionally, the process of increasing the number of valid signal samples may further include:

[0016] Comparing the valid signal samples carrying each type of microseismic event labels;

[0017] The valid signal samples corresponding to each type of microseismic event label, except for the valid signal samples corresponding to the type of microseismic event label with the largest number of samples, are upsampled respectively, and the signal samples obtained by upsampling are used as the valid signal samples corresponding to the microseismic event label, so that the number of samples corresponding to each type of microseismic event label is the same.

[0018] Optionally, obtaining valid signal samples with microseismic event labels includes:

[0019] Collecting microseismic signal samples and extracting effective signal samples from the microseismic signals;

[0020] Performing feature extraction on the effective signal sample to obtain a feature data sample;

[0021] Performing cluster analysis on the valid signal samples according to the characteristic data samples, and setting the microseismic event labels of the same type of valid signal samples that meet the first waveform condition as underground fluid movement, setting the microseismic event labels of the same type of valid signal samples that meet the second waveform condition as underground tectonic stress changes, and setting the microseismic event labels of the same type of valid signal samples that meet the third waveform condition as coal rock stratum rupture according to the cluster analysis results;

[0022] Among them, the first waveform condition is to include P wave but not S wave; the second waveform condition is to include S wave and P wave at the same time; the third waveform condition is the second largest amplitude whose difference between multiple and maximum amplitude is less than the preset difference.

[0023] Optionally, after clustering the effective signal samples according to the characteristic data samples, and before setting microseismic event labels corresponding to the effective signal samples of the same type, the method further comprises:

[0024] A normalization operation is performed on the amplitude values ​​of the valid signal samples of the same type, so as to set the microseismic event labels according to the normalized valid signal samples.

[0025] A microseismic event identification and classification device, comprising:

[0026] A feature extraction module is used to extract feature data of effective signals from the microseismic signals to be analyzed;

[0027] An event recognition module, used to identify microseismic events on the characteristic data through a pre-created neural network recognition model, and determine the microseismic events corresponding to the effective signal;

[0028] A model creation module, wherein the model creation module comprises:

[0029] A sample acquisition unit, used to obtain valid signal samples with microseismic event labels;

[0030] A sample enhancement unit, configured to process the valid signal samples according to at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending, and use the valid signal samples before and after the processing as valid signal samples to increase the number of samples of the valid signal samples;

[0031] A feature extraction unit, used to extract features from the valid signal samples to obtain feature data samples;

[0032] The model training unit is used to perform neural network training according to the characteristic data samples and the microseismic event labels carried therein to obtain the neural network recognition model.

[0033] Optionally, the model training unit is specifically used to create an LSTM network, and input the feature data samples and the earthquake event labels to perform neural network training.

[0034] Optionally, the sample enhancement unit is specifically used to compare the valid signal samples carrying each type of microseismic event label; up-sample the valid signal samples corresponding to each type of microseismic event label except the valid signal samples of the type of microseismic event label with the largest number of samples, and use the up-sampled signals as valid signal samples corresponding to the microseismic event label, so that the number of samples corresponding to each type of microseismic event label is the same.

[0035] A microseismic event identification and classification device, comprising:

[0036] Memory for storing computer programs;

[0037] A processor is used to execute the computer program to perform the steps of implementing the microseismic event identification and classification method as described in any one of the above items.

[0038] A computer-readable storage medium stores a computer program, wherein the computer program is executed by a processor to implement the steps of the microseismic event identification and classification method as described in any one of the above items.

[0039] A microseismic event identification and classification method provided by the present invention comprises extracting characteristic data of effective signals in microseismic signals to be analyzed; identifying microseismic events on the characteristic data through a pre-created neural network identification model, and determining the microseismic events corresponding to the effective signals; wherein the process of pre-creating the neural network identification model comprises: obtaining effective signal samples with microseismic event labels; processing the effective signal samples according to at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending, and taking the effective signal samples before and after the processing as effective signal samples to increase the number of effective signal samples; extracting features from the effective signal samples to obtain characteristic data samples; and training the neural network according to the characteristic data samples and the microseismic event labels carried therein to obtain the neural network identification model.

[0040] In the present application, a neural network recognition model is obtained by pre-training a neural network, and in the process of training the neural network recognition model, full consideration is given to the problem that microseismic event signals are often difficult to collect, resulting in insufficient training samples. In order to avoid overfitting of the neural network recognition model obtained through neural network training due to insufficient samples, resulting in inaccurate neural networks, the present application further adopts at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending for processing, and uses the sample data before and after processing as sample data for neural network training. On the basis of ensuring that the sample data still retains its original characteristics, the number of samples for neural network training is sufficient, thereby ensuring the accuracy of the neural network recognition model obtained through neural network training, and further ensuring the accuracy of subsequent use of the neural network model to identify the corresponding microseismic time of microseismic signals, providing accurate data basis for formation safety detection in coal mines, oil field wells, etc., which is conducive to ensuring the safety of kerosene mining.

[0041] The present application also provides a microseismic event identification and classification device, equipment and computer-readable storage medium, which have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A schematic diagram of a process flow of a microseismic event identification and classification method provided in an embodiment of the present application;

[0044] Figure 2 A structural block diagram of a microseismic event identification and classification device provided in an embodiment of the present invention;

[0045] Figure 3 A structural block diagram of a microseismic event identification and classification device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0047] like Figure 1 As shown, Figure 1 A schematic diagram of a process flow of a microseismic event identification and classification method provided in an embodiment of the present application. The microseismic event identification and classification method may include:

[0048] S11: Obtain valid signal samples with microseismic event labels in advance.

[0049] S12: Processing the valid signal samples according to at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending, and taking the valid signal samples before and after the processing as valid signal samples to increase the number of valid signal samples.

[0050] It should be noted that in order to monitor microseismic events in real time, it is generally necessary to collect microseismic signals in real time through seismic signal detection equipment. However, in fact, only valid signals collected during the time period when microseismic events occur are of analytical value.

[0051] But obviously in the actual application process, the frequency of microseismic events is relatively low, that is, they do not occur frequently. This also leads to a relatively small number of valid signal samples of microseismic signals to a certain extent, at least relative to the training of neural networks.

[0052] When training a neural network, insufficient samples may lead to overfitting of the neural network model obtained through training, which ultimately results in low accuracy of the neural network model obtained through training.

[0053] Therefore, in order to solve this problem, in this embodiment, a relatively small number of valid signal samples are processed to obtain more samples.

[0054] Specifically, the existing valid signal samples can be processed by at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending. The following will explain various ways of processing valid signal samples:

[0055] 1) Bandpass filtering can be used to increase the number without changing the label. The number of valid signal samples can be increased by performing a bandpass filtering on the valid signal samples. The bandpass filter used can be a conventional filter for filtering out noise signals, which is not limited in this embodiment.

[0056] 2) Adding Gaussian noise can be performed by calculating the original signal signal-to-noise ratio of the valid signal samples before and after bandpass filtering to obtain a reference noise ratio; on this basis, adding a sequence noise consisting of random numbers different from the reference noise ratio to the valid signal samples after bandpass filtering to obtain a new set of valid signal samples.

[0057] 3) Window Slicing (WS) randomly selects signals within a time period of fixed size but smaller than the effective signal sample duration from the effective signal samples as sample signals, thereby expanding the number of samples. Each signal slice extracted from the same waveform will be given the same label as the effective signal sample.

[0058] 4) Window Warp (WW) is a data enhancement technique unique to time series signals. It enhances time series data by accelerating or decelerating (stretching or compressing) the time series signal within the slice window. The slice window length and curvature are two parameters that affect the results of the window warping method. The relationship between the length of the time series signal after transformation and the slice length and curvature is as follows: Where N new is the length of the transformed signal, N ori is the original signal length, N is the slice length, and σ is the curvature. The transformation process when the curvature is greater than 1 is called deceleration (stretching), and the transformation process when the curvature is less than 1 is called acceleration (compression). Deceleration increases the length of the time series signal, while acceleration reduces the length of the time series signal. The window bending method generates time series signal sequences of different lengths, and the newly generated time series signal sequences can be used as valid signal samples.

[0059] The signal after the window bending transformation can also be sliced ​​again to keep it at the same length as the signal sample before the bending transformation. The combined use of window slicing and window bending methods can further increase the number of training set samples.

[0060] It should be noted that in the actual sampling process, the number of effective signal samples that can be collected for each microseismic event is often different, which, to a certain extent, causes the imbalance of effective signal samples corresponding to coal microseismic events. When the neural network is subsequently trained based on the effective signal samples of various microseismic events, the imbalance of sample data will also cause the overfitting problem of the neural network recognition model obtained through training.

[0061] To this end, in an optional embodiment of the present application, it may further include:

[0062] Compare the valid signal samples carrying each type of microseismic event label;

[0063] The valid signal samples corresponding to each type of microseismic event label except the valid signal samples corresponding to the type of microseismic event label with the largest number of samples are upsampled respectively, and the signal samples obtained by upsampling are used as the valid signal samples of the corresponding microseismic event label, so that the number of samples corresponding to each type of microseismic event label is the same.

[0064] Upsampling is to copy the effective signal samples corresponding to one or several categories of microseismic events with fewer effective signal samples, so that the number of effective signal samples corresponding to each microseismic event is basically the same, so as to enhance the balance between the effective signal samples of each category of microseismic events.

[0065] It is understandable that when sampling the valid signal samples of each microseismic event, the number of samples that need to be copied is determined based on the original number of samples, as long as the number of samples of all categories is basically the same in the end.

[0066] In addition, upsampling can be performed by performing any of the above-mentioned bandpass filtering, adding Gaussian noise, window slicing, window bending, or centralized processing on the effective signal samples corresponding to each microseismic event, and then comparing the effective signal samples of each microseismic event, and upsampling the original effective signal samples that are insufficient to copy enough effective signal samples.

[0067] Furthermore, in order to avoid overfitting of the model due to repeated samples during upsampling, some slight disturbances, i.e. random noise, can be added to the newly generated samples after upsampling, thereby achieving the purpose of enhancing the balance of the data set without overfitting.

[0068] Of course, in practical applications, it is not excluded to achieve a balance in the number of samples in the process of bandpass filtering, window bending, window slicing, etc. of the effective signal samples. For example, when performing window slicing or window bending processing on the effective signal samples of microseismic events with a large number of samples, a smaller number of slicing and bending processing can be selected to make the increment of the effective signal samples relative to other analog microseismic signals. There are also similar processing methods, which are not listed one by one in this application.

[0069] S13: Extract features from valid signal samples to obtain feature data samples.

[0070] It should be noted that, feature extraction is performed on valid signal samples, and specifically, multiple feature data samples can be extracted from the feature data of valid signal samples including total energy, average energy, total entropy, average entropy, total ringing, average ringing, rise time, duration, main frequency, maximum amplitude absolute value, P wave S wave time difference, and P wave S wave amplitude ratio.

[0071] Each characteristic data is described in detail below.

[0072] 1) About total energy and average energy: The square of the amplitude corresponding to the effective signal segment is used to represent the energy. The average energy is the arithmetic mean of the total energy Among them, N is the number of sampling points in the effective signal segment, A i Indicates the amplitude value of the effective signal at the i-th sampling point.

[0073] 2) About total entropy and average entropy: Entropy is a parameter used to describe the disorder of things. The larger the entropy, the more chaotic it is. Entropy represents the uncertainty of a random variable. For the amplitude value of an effective signal, which is a random variable, its possible values ​​are x1, x2, x3...x m , the corresponding probabilities of each value are p1, p2, p3, ..., p m , then the expressions of total entropy and average entropy are:

[0074] 3) About total ringing and average ringing: The overshoot and continuous ripples generated when the effective signal oscillates back and forth near the steady state are called ringing effects, which are physical quantities that characterize signal stability. For the effective signal of microseismic events, the number of ringing effects, that is, the total ringing, can be expressed by the number of zero crossings of the effective signal: L = length (find (A i A i+1 ≤0)); the corresponding average ringing can be A i and A i+1 is the amplitude of two adjacent sampling points, and N is the number of sampling points in the valid signal segment.

[0075] 4) Rise time and duration: The time interval between the first arrival time of the sampling point at the starting point of the valid signal segment and the valid signal sampling point with the maximum amplitude is the rise time; the duration corresponding to the valid signal time period is the duration.

[0076] 5) Maximum absolute value of amplitude and main frequency: Perform effective value calculation on the amplitude values ​​of each effective signal, and the maximum absolute value obtained is also the maximum positive and negative absolute value. The frequency value corresponding to the signal with the maximum absolute value of amplitude is called the main frequency: F max =f(|A max |).

[0077] 6) P wave and S wave time difference:

[0078] There are two main types of seismic waves, one is surface waves and the other is body waves. Surface waves are only transmitted on the surface, while body waves can pass through the interior of the earth. Body waves: They are transmitted inside the earth and are divided into P waves and S waves. P waves: P stands for primary or compression, and is a longitudinal wave. The direction of particle vibration is parallel to the direction of wave advance. Among all seismic waves, it has the fastest forward speed and arrives first. P waves can be transmitted in solids, liquids or gases. S waves: S means secondary or shear. The forward speed is second only to P waves. The direction of particle vibration is perpendicular to the direction of wave advance. It is a transverse wave. S waves can only be transmitted in solids and cannot pass through the liquid outer core. By taking advantage of the different transmission speeds of P waves and S waves and the travel time difference between the two, simple earthquake positioning can be performed.

[0079] In this embodiment, the P-wave S-wave time difference, i.e., the time difference between the first arrival of the S-wave and the P-wave, can be calculated by the formula It is determined that st(S) represents the first arrival time of the S wave, and st(P) represents the first arrival time of the P wave.

[0080] P wave to S wave amplitude ratio: that is, the absolute value of the ratio of the maximum amplitude of the S wave to the maximum amplitude of the P wave, which can be calculated by the formula OK, A max (S) represents the maximum amplitude of the S wave, A max (P) indicates the maximum amplitude of the P wave.

[0081] S14: Perform neural network training based on the characteristic data samples and the microseismic event labels they carry to obtain a neural network recognition model.

[0082] Optionally, the neural network used in this embodiment may be an LSTM network; an LSTM network may be created, and the characteristic data samples obtained above and the corresponding microseismic event labels may be used as input for computer learning training to obtain a neural network recognition model.

[0083] S15: Extract characteristic data of effective signals from the microseismic signals to be analyzed.

[0084] It is understandable that in this embodiment, when extracting feature data of effective signals in the microseismic signals to be analyzed, the feature data samples extracted for sample training should be of the same type of feature data, which will not be described in detail.

[0085] S16: Identify microseismic events of characteristic data through a pre-created neural network recognition model to determine the microseismic events corresponding to the effective signals.

[0086] In the present application, neural network training is performed in advance based on effective signal samples corresponding to the microseismic time, and a neural network recognition model is obtained that can analyze and identify the effective signal corresponding to the occurrence of a microseismic event, and determine which type of microseismic event the effective signal specifically belongs to. In this way, when the effective signal of the microseismic event is subsequently acquired, the neural network recognition model can be directly used to realize rapid and accurate recognition of the microseismic event, which is beneficial to improving the efficiency and accuracy of microseismic event detection.

[0087] In addition, in the process of creating the neural network recognition model in the present application, the low frequency of occurrence of microseismic events themselves is taken into consideration, which makes it difficult to obtain the corresponding effective signals, resulting in insufficient signal samples for neural network training, which can easily cause the trained model to have overfitting and low recognition accuracy. After collecting a limited number of effective signal samples, the effective signal samples are processed to increase the number of samples, thereby providing enough samples for neural network training, ensuring the accuracy of the neural network recognition model obtained by training, and then ensuring the accuracy of the neural network recognition model in identifying microseismic events corresponding to the effective signals obtained by real-time collection.

[0088] Based on the above discussion, after obtaining valid signal samples in the microseismic signal, each valid signal sample needs to carry the label of its corresponding microseismic event. Although the valid signal samples in the microseismic signal obtained by collection have the problem of insufficient samples for neural network training, it is obviously a very large workload to rely entirely on manual analysis of the corresponding microseismic event labels of the valid signal samples one by one. Therefore, in an optional embodiment of the present application, the process of determining the label corresponding to each valid signal sample may include:

[0089] S21: Collect microseismic signal samples and extract valid signal samples from the microseismic signals.

[0090] S22: Extract features from valid signal samples to obtain feature data samples.

[0091] S23: Perform cluster analysis on the valid signal samples according to the characteristic data samples, and divide each valid signal sample into different cluster categories.

[0092] For example, the agglomerative hierarchical clustering analysis algorithm can be used to classify valid signals. Agglomerative hierarchical clustering analysis is a bottom-up strategy that considers the feature data corresponding to each valid signal sample as a cluster (one class, N classes in total), performs similarity calculations or Mahalanobis distance calculations on the feature data of each valid signal sample, finds two valid signal segments with the highest similarity or the smallest distance, and merges them into the same cluster. This process is then iterated, and each cluster is continuously merged until all objects are in one cluster or a certain iteration termination condition is met. For a clustering operation containing N valid signal segments, the sum of the number of object types m and the number of clustering times n in the final clustering result is the same as the number of valid signal segments.

[0093] S24: setting the microseismic event labels of the valid signal samples of the same cluster category that meet the first waveform condition as underground fluid movement.

[0094] S25: setting the microseismic event labels of the valid signal samples of the same cluster category that meet the second waveform condition as underground tectonic stress changes.

[0095] S26: setting the microseismic event labels of the valid signal samples of the same cluster category that meet the third waveform condition as coal stratum rupture.

[0096] The first waveform condition is to include the P wave but not the S wave.

[0097] For the seismic signal waveform corresponding to underground fluid movement, the waveform characteristics are that the maximum amplitude value of the signal is significantly higher than other amplitude values, and the rise time and duration corresponding to the effective signal segment are shorter, and the total number of ringing is also smaller.

[0098] The second waveform condition is to include both the S wave and the P wave.

[0099] The seismic signal waveform corresponding to the change of underground tectonic stress contains two obvious peaks, namely the P wave and the S wave. The rise time of the P wave and the S wave is short, the duration of the effective signal segment is long, and the total number of ringings is small.

[0100] The third waveform condition is that the difference between the plurality of amplitudes and the maximum amplitude is smaller than the second largest amplitude of the preset difference.

[0101] For the seismic signal waveform corresponding to the coal-rock rupture, there are multiple amplitude values ​​close to the maximum amplitude value, and the signal rise time and duration are long, and the total number of ringing is also large.

[0102] It can be understood that, for the valid signal samples of the same cluster category that meet the first waveform condition, it is not required that all valid signal samples in the category fully meet the first waveform condition, because in actual applications, some valid signal samples are signal samples generated by underground fluid movement, but their waveform characteristics may not be particularly obvious in the waveform manifestation of S waves and P waves. Therefore, as long as most of the valid signal samples of the same category or more than a certain proportion of the valid signal samples meet the first waveform, it can be considered that the valid signal samples of the category belong to the signal samples corresponding to the underground fluid movement.

[0103] Similarly, for valid signal samples generated by changes in underground structural stress, only a certain proportion of valid sample signals are required to meet the second waveform condition; and among the valid signal samples generated by coal seam rupture, only most of the samples are required to meet the third waveform condition.

[0104] It should be noted that after clustering analysis is performed on each effective signal segment to divide it into categories, in order to better analyze and identify the waveform characteristics of the effective signal segments corresponding to each category, so as to determine the seismic events to which the effective signal segments of the same category belong, the amplitude values ​​of the effective signals in each effective signal segment can be normalized so that the amplitude values ​​of the effective signals in each effective signal segment are converted to the same order of magnitude, which facilitates the comparison of the effective signals between each effective signal segment.

[0105] The following is an introduction to a microseismic event identification and classification device provided in an embodiment of the present invention. The microseismic event identification and classification device described below and the microseismic event identification and classification method described above can be referenced to each other.

[0106] Figure 2 The structural block diagram of the microseismic event identification and classification device provided by the embodiment of the present invention is shown in FIG. Figure 2 The microseismic event identification and classification device may include:

[0107] A feature extraction module 100 is used to extract feature data of effective signals from the microseismic signals to be analyzed;

[0108] An event recognition module 200 is used to recognize microseismic events on the characteristic data through a pre-created neural network recognition model, and determine the microseismic events corresponding to the effective signal;

[0109] The model creation module 300 includes:

[0110] The sample collection unit 301 is used to obtain valid signal samples with microseismic event labels;

[0111] The sample enhancement unit 302 is used to process the valid signal samples according to at least one method of bandpass filtering, adding Gaussian noise, window slicing, and window bending, and use the valid signal samples before and after the processing as valid signal samples to increase the number of samples of the valid signal samples;

[0112] A feature extraction unit 303 is used to extract features from the valid signal samples to obtain feature data samples;

[0113] The model training unit 304 is used to perform neural network training according to the characteristic data samples and the microseismic event labels carried therein to obtain the neural network recognition model.

[0114] In an optional embodiment of the present application, the model training unit 304 is specifically used to create an LSTM network, and input the feature data sample and the earthquake event label to perform neural network training.

[0115] In an optional embodiment of the present application, the sample enhancement unit 302 is specifically used to compare the valid signal samples carrying each type of microseismic event label; up-sample the valid signal samples corresponding to each type of microseismic event label except the valid signal samples of the type of microseismic event label with the largest number of samples, and use the up-sampled signals as valid signal samples corresponding to the microseismic event label, so that the number of samples corresponding to each type of microseismic event label is the same.

[0116] In an optional embodiment of the present application, the sample acquisition unit 301 is specifically used to acquire microseismic signal samples and extract valid signal samples from the microseismic signals;

[0117] Performing feature extraction on the effective signal sample to obtain a feature data sample;

[0118] A cluster analysis is performed on the valid signal samples according to the characteristic data samples, and according to the cluster analysis results, the microseismic event labels of the same type of valid signal samples that meet the first waveform condition are set as underground fluid movement, the microseismic event labels of the same type of valid signal samples that meet the second waveform condition are set as underground tectonic stress changes, and the microseismic event labels of the same type of valid signal samples that meet the third waveform condition are set as coal rock layer rupture; wherein, the first waveform condition is to include P waves but not S waves; the second waveform condition is to include S waves and P waves at the same time; the third waveform condition is the second largest amplitude whose difference between multiple and the maximum amplitude is less than a preset difference.

[0119] In an optional embodiment of the present application, the sample acquisition unit 301 is specifically configured to perform a normalization operation on the amplitude values ​​of the valid signal samples of the same type, so as to set the microseismic event label according to the normalized valid signal samples.

[0120] The microseismic event identification and classification device of this embodiment is used to implement the aforementioned microseismic event identification and classification method. Therefore, the specific implementation method of the feature extraction device can be seen in the embodiment part of the feature extraction method in the previous text. Its specific implementation method can refer to the description of the corresponding each part embodiment, which will not be repeated here.

[0121] The present application also provides an embodiment of a microseismic event identification and classification device, Figure 3 A structural block diagram of a microseismic event identification and classification device provided in an embodiment of the present invention, the device may include:

[0122] A memory 11, used for storing computer programs;

[0123] The processor 12 is used to execute the computer program to perform the steps of implementing the microseismic event identification and classification method as described in any one of the above items.

[0124] The steps of the microseismic event identification and classification method executed by the processor may include:

[0125] Extract characteristic data of effective signals from microseismic signals to be analyzed;

[0126] Identify microseismic events on the characteristic data using a pre-created neural network recognition model to determine the microseismic events corresponding to the effective signal;

[0127] The process of pre-creating the neural network recognition model includes:

[0128] Obtain valid signal samples with microseismic event labels;

[0129] Processing the valid signal samples according to at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending, and using the valid signal samples before and after the processing as valid signal samples to increase the number of valid signal samples;

[0130] Performing feature extraction on the effective signal sample to obtain a feature data sample;

[0131] A neural network training is performed based on the characteristic data samples and the microseismic event labels carried therein to obtain the neural network recognition model.

[0132] The present application also provides a computer-readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps of the microseismic event identification and classification method as described in any one of the above items.

[0133] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment that includes a series of elements are inherent to the elements. In the absence of more restrictions, the elements limited by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or equipment that includes the elements. In addition, the above-mentioned technical solution provided in the embodiment of the present application is consistent with the corresponding technical solution in the prior art in principle, and the part is not described in detail, so as not to repeat too much.

[0134] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

Claims

1. A method for identifying and classifying microseismic events, characterized in that: include: Extract characteristic data of effective signals from microseismic signals to be analyzed; Identify microseismic events on the characteristic data using a pre-created neural network recognition model to determine the microseismic events corresponding to the effective signal; The process of pre-creating the neural network recognition model includes: Obtain valid signal samples with microseismic event labels; Processing the valid signal samples according to at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending, and using the valid signal samples before and after the processing as valid signal samples to increase the number of valid signal samples; Performing feature extraction on the effective signal sample to obtain a feature data sample; Performing neural network training according to the characteristic data samples and the microseismic event labels carried therein to obtain the neural network recognition model; The step of obtaining a valid signal sample with a microseismic event label comprises: Collecting microseismic signal samples and extracting effective signal samples from the microseismic signals; Performing feature extraction on the effective signal sample to obtain a feature data sample; Performing cluster analysis on the valid signal samples according to the characteristic data samples, and setting the microseismic event labels of the same type of valid signal samples that meet the first waveform condition as underground fluid movement, setting the microseismic event labels of the same type of valid signal samples that meet the second waveform condition as underground tectonic stress changes, and setting the microseismic event labels of the same type of valid signal samples that meet the third waveform condition as coal rock stratum rupture according to the cluster analysis results; Among them, the first waveform condition is to include P wave but not S wave; the second waveform condition is to include S wave and P wave at the same time; the third waveform condition is the second largest amplitude whose difference between multiple and maximum amplitude is less than the preset difference.

2. The microseismic event identification and classification method according to claim 1, characterized in that: Performing neural network training according to the characteristic data sample and the earthquake event label carried therein, including: An LSTM network is created, and the characteristic data samples and the earthquake event labels are input to perform neural network training.

3. The microseismic event identification and classification method according to claim 1, characterized in that: The process of increasing the number of valid signal samples includes: Comparing the valid signal samples carrying each type of microseismic event labels; The valid signal samples corresponding to each type of microseismic event label, except for the valid signal samples corresponding to the type of microseismic event label with the largest number of samples, are upsampled respectively, and the signal samples obtained by upsampling are used as the valid signal samples corresponding to the microseismic event label, so that the number of samples corresponding to each type of microseismic event label is the same.

4. The microseismic event identification and classification method according to claim 1, characterized in that: After clustering the effective signal samples according to the characteristic data samples and before setting microseismic event labels corresponding to the effective signal samples of the same type, the method further includes: A normalization operation is performed on the amplitude values ​​of the valid signal samples of the same type, so as to set the microseismic event labels according to the normalized valid signal samples.

5. A microseismic event identification and classification device, characterized in that: include: A feature extraction module is used to extract feature data of effective signals from the microseismic signals to be analyzed; An event recognition module, used to identify microseismic events on the characteristic data through a pre-created neural network recognition model, and determine the microseismic events corresponding to the effective signal; A model creation module, wherein the model creation module comprises: A sample acquisition unit, used to obtain valid signal samples with microseismic event labels; A sample enhancement unit, configured to process the valid signal samples according to at least one of bandpass filtering, adding Gaussian noise, window slicing, and window bending, and use the valid signal samples before and after the processing as valid signal samples to increase the number of the valid signal samples; A feature extraction unit, used to extract features from the valid signal samples to obtain feature data samples; A model training unit, used to perform neural network training according to the characteristic data samples and the microseismic event labels carried therein to obtain the neural network recognition model; The sample acquisition unit acquires microseismic signal samples and extracts valid signal samples from the microseismic signals; performs feature extraction on the valid signal samples to obtain feature data samples; performs cluster analysis on the valid signal samples according to the feature data samples, and sets the microseismic event labels of the same type of valid signal samples that meet the first waveform condition as underground fluid movement, sets the microseismic event labels of the same type of valid signal samples that meet the second waveform condition as underground tectonic stress changes, and sets the microseismic event labels of the same type of valid signal samples that meet the third waveform condition as coal rock stratum rupture according to the cluster analysis results; Among them, the first waveform condition is to include P wave but not S wave; the second waveform condition is to include S wave and P wave at the same time; the third waveform condition is the second largest amplitude whose difference between multiple and maximum amplitude is less than the preset difference.

6. The microseismic event identification and classification device according to claim 5, characterized in that: The model training unit is specifically used to create an LSTM network, and input the characteristic data samples and the earthquake event labels to perform neural network training.

7. The microseismic event identification and classification device according to claim 5, characterized in that: The sample enhancement unit is specifically used to compare the valid signal samples carrying each type of microseismic event label; up-sample the valid signal samples corresponding to each type of microseismic event label except the valid signal samples of the type of microseismic event label with the largest number of samples, and use the up-sampled signals as the valid signal samples corresponding to the microseismic event label, so that the number of samples corresponding to each type of microseismic event label is the same.

8. A microseismic event identification and classification device, characterized in that: include: Memory for storing computer programs; A processor is used to execute the computer program to perform the steps of implementing the microseismic event identification and classification method as described in any one of claims 1 to 4.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the microseismic event identification and classification method according to any one of claims 1 to 4.

Images