Automatic recognition system of seismic wave signals based on machine learning

Abstract

The invention discloses an automatic recognition system for seismic wave signals based on machine learning, and specifically relates to the technical field of machine learning, comprising a preliminary filtering module, a spectrum analysis module, a phase picking module, a feature analysis module, and a classification recognition module. The preliminary filtering module performs preliminary processing and screening on seismic wave signals, processes signals in segments and calculates their energy. The spectrum analysis module extracts spectrum features of seismic signals by using Fourier transform. The phase picking module distinguishes seismic signals from non-seismic event signals by calculating the degree of low-frequency background interference and a trigger algorithm. The feature analysis module extracts waveform features from original seismic records and transmits them to the classification recognition module. The classification recognition module establishes a model using a classification algorithm, analyzes and recognizes the extracted features, separates and recognizes seismic signals from non-seismic event signals by using low-frequency signal features and a trigger algorithm, avoids recognition failures caused by low-frequency signal stacking, and improves the accuracy and timeliness of seismic signal recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and more specifically, to an automatic recognition system for seismic wave signals based on machine learning. Background Art

[0002] Automatic identification of seismic wave signals based on machine learning is a technical method that uses machine learning technology to classify and analyze seismic wave signals. This method trains the machine learning model to enable it to automatically identify and classify different types of seismic wave signals and noise signals. Seismic waves are the energy generated when an earthquake occurs and propagate in the form of waves. They mainly include P waves, S waves and surface waves, among which P waves are the fastest seismic waves, followed by S waves, and surface waves propagate the slowest. Automatic detection and classification of seismic wave signals through machine learning algorithms without human intervention is very important for real-time earthquake monitoring and early warning.

[0003] The duration of seismic signals is usually longer, generally ranging from a few seconds to a few minutes, while the duration of non-seismic events is usually shorter. Non-seismic events, such as traffic vibrations, are generally short-term and instantaneous. The current automatic identification of seismic wave signals and the study of noise signals are usually limited to earthquakes with deeper sources and more complex signal waveforms, because the seismic noise background with shallower sources is usually simpler and easier to filter. If there is interference from non-seismic signals during the continuous process of high-energy deep-source earthquakes, it will often affect the identification of seismic signals, making it difficult to judge the specific seismic signal waveform, which in turn creates obstacles to earthquake early warning and research.

[0004] In order to solve the above defects, a technical solution is now proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a seismic wave signal automatic recognition system based on machine learning to address the deficiencies in the background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solutions: a seismic wave signal automatic recognition system based on machine learning, comprising a preliminary filtering module, a spectrum analysis module, a seismic phase picking module, a feature analysis module, and a classification recognition module;

[0007] The preliminary filtering module is used to preliminarily process and filter the seismic wave signal, process the seismic signal in segments, and calculate the signal energy;

[0008] The spectrum analysis module is used to analyze the frequency characteristics of the signal, perform frequency analysis on the segmented window of the seismic signal, and extract the spectrum characteristics of the seismic signal through the Fourier transform method;

[0009] The seismic phase picking module is used to calculate the low-frequency background interference degree of the seismic signal, and combined with the trigger algorithm, distinguish and judge the seismic signal and the non-seismic event signal, and eliminate the non-seismic event signal;

[0010] The feature analysis module is used to extract the features of seismic waveforms from the original seismic records and transmit them to the classification and recognition module;

[0011] The classification and recognition module is used to analyze and identify the extracted features, use the classification algorithm to build a model, and classify and judge the seismic signals.

[0012] Preferably, the method for segmenting the seismic signal is:

[0013] For the square sum formula of signal energy, for a given time window x(t), its energy is expressed as E=∑ t N ＝1 x(t) 2 , where E is the energy signal intensity, that is, the total energy accumulation of the signal in a certain period of time, x(t) is the amplitude value of the signal at time t, indicating the instantaneous value of the signal at a certain moment, t is the sampling time of the signal, t varies from 1 to N, representing all sampling points in the time window, N is the total number of sampling points in the time window, indicating the total number of sampling points contained in the selected time window length, and N is greater than or equal to 1;

[0014] The root mean square formula of the signal energy is: Where RMS is the root mean square value, which is the square root of the energy of the signal in the selected time window, and represents the effective value of the signal in AC signal analysis.

[0015] Preferably, the method for calculating the low-frequency background interference degree of the seismic signal is:

[0016] The low-frequency background interference degree of the seismic signal is evaluated by calculating the low-frequency background complexity of the signal, and the low-frequency background frequency critical value is set. When the signal frequency is lower than the low-frequency background frequency critical value, the low-frequency background complexity is calculated using the signal window and the background window. The calculation expression is: Where SN is the low-frequency background complexity, s(t) is the waveform data of the signal window, n(t) is the waveform data of the background window, N1 is the number of data points in the signal window, s(t i ) is the waveform value at the i-th time point in the signal window, N2 is the number of data points in the background window, n(t i ) is the waveform value at the i-th time point in the background window.

[0017] Preferably, the method of combining the low-frequency background interference degree of the seismic signal with the trigger algorithm is:

[0018] Calculate the short-time average energy of the signal, the expression is In the formula, x(i) is the amplitude value of the signal at time point i, STA(t) is the short-term average value at time t, which represents the average energy of the signal in a shorter time window, and N STA is the length of the short-time average window, that is, the number of sample points in the time window used to calculate the STA value;

[0019] Calculate the long-term average energy of the signal, the expression is Where LTA(t) is the long-term average at time t, which represents the average energy of the signal in a longer time window, and N LTA is the length of the long-term average window, that is, the number of sample points in the time window used to calculate the LTA value;

[0020] By setting the trigger threshold Th, potential seismic signals are identified and determined. The calculation expression of the trigger value Tr is: If the trigger value Tr is greater than or equal to the trigger threshold Th, the credibility of seismic signal recognition is high and the characteristics of the seismic wave are more obvious; if the trigger value Tr is less than the trigger threshold Th, the credibility of seismic signal recognition is low and the characteristics of the seismic wave are less obvious.

[0021] Preferably, the method for distinguishing and judging seismic signals from non-seismic event signals and eliminating non-seismic event signals is:

[0022] When the trigger value Tr is greater than or equal to the trigger threshold Th, the signal is highly reliable as a seismic signal, the signal is marked as a valid signal, and the valid signal waveform is stored for analysis;

[0023] When the trigger value Tr is less than the trigger threshold Th, the signal is a seismic signal with low credibility, and the signal is marked as a background interference signal, which is then screened out.

[0024] Preferably, the features extracted from the original seismic records for identifying and classifying seismic waveforms include time domain features, frequency domain features, wavelet transform features, time-frequency domain joint features, high-order statistical features, and morphological features.

[0025] Preferably, the method of analyzing and identifying the extracted features, using the classification algorithm to establish a model, and classifying and judging the seismic signals is:

[0026] The extracted features are standardized, and the classification algorithm is used to train the model. The data set is divided into a training set and a test set. The training set data is used to train the model based on the selected classification algorithm. The performance of the model is evaluated through k-fold cross validation to prevent overfitting. The F1 value of the model is calculated, the ROC curve is plotted and the AUC value is calculated. The model usable thresholds of the F1 value and the AUC value of the ROC curve are preset. Only when the F1 value of the model and the AUC value of the ROC curve simultaneously meet the model usable threshold, the model enters the actual application stage, and the trained model is used to classify real-time seismic signals and identify and judge potential non-seismic event signals in continuous deep-source seismic signals.

[0027] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0028] During the continuous signal process of deep-source earthquakes, the seismic signal and the non-seismic event signal can be separated and identified by analyzing the characteristics of the low-frequency signal and combining it with the trigger algorithm. This can effectively extract valid data that can be used for earthquake early warning and earthquake research, avoid the failure of seismic signal recognition caused by the stacking of low-frequency signals, accurately pick up the fluctuation changes of the seismic waveform, and further improve the accuracy and timeliness of seismic signal recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0030] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Example 1

[0033] See also Figure 1 As shown, the present invention is an automatic identification system for seismic wave signals based on machine learning, including a preliminary filtering module, a spectrum analysis module, a seismic phase picking module, a feature analysis module, and a classification identification module;

[0034] The preliminary filtering module is used to preliminarily process and filter the seismic wave signal, process the seismic signal in segments, and calculate the signal energy;

[0035] The spectrum analysis module is used to analyze the frequency characteristics of the signal, perform frequency analysis on the segmented window of the seismic signal, and extract the spectrum characteristics of the seismic signal through the Fourier transform method;

[0036] The seismic phase picking module is used to calculate the low-frequency background interference degree of the seismic signal, and combined with the trigger algorithm, distinguish and judge the seismic signal and the non-seismic event signal, and eliminate the non-seismic event signal;

[0037] The feature analysis module is used to extract the features of seismic waveforms from the original seismic records and transmit them to the classification and recognition module;

[0038] The classification and recognition module is used to analyze and identify the extracted features, use the classification algorithm to build a model, and classify and judge the seismic signals.

[0039] The preliminary filtering module can perform preliminary processing and screening of seismic wave signals in order to identify possible non-seismic event signals during the duration of deep-source seismic signals. This step is crucial because it can significantly reduce the amount of calculation for subsequent analysis and improve the accuracy of identification. The specific steps of preliminary filtering are:

[0040] Signal segmentation: Divide the continuous seismic wave signal into multiple short time windows according to time. The short time window time can be one second or several seconds, which is specifically set by professional and technical personnel in this field. The short time window segmentation can capture the changes of the signal on the time axis. It should be noted that the selection of the time window length depends on the time characteristics of the target event. Generally, the time window length of 1-10 seconds is more commonly used, which can take into account the characteristics of seismic waves and non-seismic event signals;

[0041] Amplitude envelope calculation: Calculate the amplitude envelope of the signal in each time window, obtained through Hilbert transform or other methods. The envelope reflects the amplitude change of the signal. Through Hilbert transform, the instantaneous amplitude of the signal, that is, the amplitude envelope, can be obtained;

[0042] Energy analysis: Calculate the energy of the signal in each time window. The calculation method can be the sum of squares or the root mean square. Seismic events usually have higher energy, while the energy of non-seismic events is relatively low or varies significantly.

[0043] For the square sum formula of signal energy, for a given time window x(t), its energy is expressed as In the formula, E is the energy signal intensity, that is, the total energy accumulation of the signal in a certain period of time, x(t) is the amplitude value of the signal at time t, indicating the instantaneous value of the signal at a certain moment, t is the sampling time of the signal, t varies from 1 to N, representing all sampling points in the time window, N is the total number of sampling points in the time window, indicating the total number of sampling points contained in the selected time window length, and N is greater than or equal to 1;

[0044] The root mean square formula of the signal energy is: In the formula, RMS is the root mean square value, which is the square root of the energy of the signal in the selected time window, and represents the effective value of the signal in AC signal analysis;

[0045] The spectrum analysis module can analyze the frequency characteristics of the signal. In seismic wave signal recognition, spectrum analysis can effectively distinguish seismic signals from non-seismic event signals. Seismic signals and non-seismic event signals have different spectrum characteristics. The specific steps are as follows:

[0046] Signal preprocessing: remove the trend component in the signal to avoid the interference of low-frequency components, which can be achieved by subtracting the mean of the signal or applying a high-pass filter. By filtering, high-frequency and low-frequency noise in the signal are removed to improve the signal-to-noise ratio.

[0047] Select analysis window: Divide the continuous signal into multiple short time windows. The length of the time window depends on the analysis target and signal characteristics. Apply a window function to each time window to reduce spectrum leakage. The window function can smooth the boundaries of the time window and improve the accuracy of spectrum estimation. Commonly used window functions include Hamming window, Hanning window, Blackman window, etc.

[0048] Short-time Fourier transform: Perform a short-time Fourier transform on the signal in each time window to obtain a time-frequency representation. The short-time Fourier transform calculates the local Fourier transform at different time points through a sliding window function. The expression is Where X(t,f) is the result of short-time Fourier transform, which is the complex value representation of the signal at time t and frequency f. It represents the amplitude and phase information at time t and frequency f. x(τ) is the original signal before the transform. The time variable is τ, which represents the amplitude value of the signal at time τ. w(t-τ) is the window function, which is a weight function used to window the signal at time τ. The window function is usually a smooth function, such as Hamming window, Hanning window or Gaussian window, which can localize the signal for local Fourier transform. t is the time offset parameter, which represents the time point where the center of the window function is located. τ is the integral variable, which represents the original time axis of the signal. -j2πfτIt is a complex exponential function and is the core part of Fourier transform, where j is an imaginary unit, 2πfτ is the phase change of frequency component f at time τ, and dτ is the integral variable, which means integrating τ. Short-time Fourier transform can provide the time-frequency representation of the signal, helping to analyze the change of the frequency of the signal over time. In seismic signal analysis, short-time Fourier transform can identify different types of seismic waves and their frequency characteristics, and distinguish between seismic signals and non-seismic event signals.

[0049] Spectral feature extraction: Calculate the power spectrum density of the signal to represent the energy distribution of each frequency component. The calculation formula for power spectrum density is: Where P(f) is the power spectrum density, which indicates the power intensity of the signal at frequency f, reflecting the energy distribution of the signal on the frequency component; X(f) is the Fourier transform result of the signal, which indicates the amplitude and phase of the signal at frequency f;

[0050] The seismic phase picking module can automatically identify and extract the arrival times of different types of seismic waves from seismic signal records, which is used to distinguish and judge seismic signals and non-seismic event signals.

[0051] Calculate the short-time average energy of the signal, the expression is In the formula, x(i) is the amplitude value of the signal at time point i, STA(t) is the short-term average value at time t, which represents the average energy of the signal in a shorter time window, and N STA is the length of the short-time average window, that is, the number of sample points in the time window used to calculate the STA value;

[0052] Calculate the long-term average energy of the signal, the expression is Where LTA(t) is the long-term average at time t, which represents the average energy of the signal in a longer time window, and N LTA is the length of the long-term average window, that is, the number of sample points in the time window used to calculate the LTA value;

[0053] The low-frequency background interference degree of the seismic signal is evaluated by calculating the low-frequency background complexity of the signal, and the low-frequency background frequency critical value is set. When the signal frequency is lower than the low-frequency background frequency critical value, the low-frequency background complexity is calculated using the signal window and the background window. The calculation expression is: Where SN is the low-frequency background complexity, s(t) is the waveform data of the signal window, n(t) is the waveform data of the background window, N1 is the number of data points in the signal window, s(t i ) is the waveform value at the i-th time point in the signal window, N2 is the number of data points in the background window, n(t i ) is the waveform value at the i-th time point in the background window;

[0054] By setting the trigger threshold Th, potential seismic signals are identified and determined. The calculation expression of the trigger value Tr is: If the trigger value Tr is greater than or equal to the trigger threshold Th, the earthquake signal recognition credibility is high, the characteristics of the earthquake wave are more obvious, and the earthquake warning is more timely and accurate; if the trigger value Tr is less than the trigger threshold Th, the earthquake signal recognition credibility is low, the characteristics of the earthquake wave are less obvious, and the earthquake warning is not timely or accurate;

[0055] When the trigger value Tr is greater than or equal to the trigger threshold Th, the signal is highly reliable as a seismic signal, the signal is marked as a valid signal, and the valid signal waveform is stored for analysis;

[0056] When the trigger value Tr is less than the trigger threshold Th, the signal is a seismic signal with low credibility, and the signal is marked as a background interference signal, which is then screened out.

[0057] The feature analysis module extracts important features for identifying and classifying seismic waveforms from raw seismic records. The verified features include the following types:

[0058] Time domain feature extraction:

[0059] The arrival time includes the P-wave arrival time and the S-wave arrival time. The initial arrival time of the P-wave is determined by the trigger algorithm. The initial arrival time of the S-wave is determined by the trigger algorithm in combination with the propagation characteristics of the seismic waves.

[0060] The amplitude includes peak amplitude and RMS amplitude. It measures the maximum amplitude of P waves and S waves, can estimate the earthquake scale, calculate the RMS amplitude in the P wave and S wave band, and reflect the signal energy.

[0061] The initial motion polarity assists in focal mechanism analysis by recording the positive and negative polarity of the initial amplitude of the P wave;

[0062] Phase duration is used for source characteristics analysis by measuring the duration from the arrival time of the P wave to the arrival time of the S wave;

[0063] Frequency domain feature extraction:

[0064] The power spectral density calculates the energy distribution of different frequency components through short-time Fourier transform or fast Fourier transform;

[0065] Spectral features are used to distinguish different types of seismic waves by extracting features such as spectrum peaks and frequency bandwidth;

[0066] Band energy is used to distinguish between seismic waves and noise by calculating the total energy within a specific frequency range;

[0067] The main frequency is the frequency component with the largest identification energy, reflecting the main oscillation frequency of the signal;

[0068] Wavelet transform feature extraction:

[0069] Continuous wavelet transform includes time-frequency representation and wavelet coefficients. CWT is used to obtain the energy distribution of signals at different scales and time positions, extract seismic wave characteristics, and identify the characteristics of different seismic phases by analyzing the changes in wavelet transform coefficients.

[0070] Discrete wavelet transform includes multi-resolution analysis and signal reconstruction. The signal is decomposed into different frequency bands through DWT, and the features in each frequency band are extracted for seismic phase identification. The signal of a specific frequency band is reconstructed through inverse transformation, and the seismic phase characteristics of the specific frequency band are further analyzed.

[0071] Joint feature extraction in time-frequency domain:

[0072] Short-time Fourier transform combines time and frequency information to obtain the time-frequency representation of the signal.

[0073] Hilbert-Huang transform extracts the features of nonlinear and non-stationary signals through empirical mode decomposition and Hilbert transform;

[0074] High-order statistical feature extraction:

[0075] The skewness can reflect the symmetry of the signal amplitude distribution, and the kurtosis can reflect the sharpness of the signal amplitude distribution;

[0076] Morphological feature extraction:

[0077] Signal morphological analysis is to extract the morphological features of the signal by using morphological operations, and the envelope is to extract the main morphological features of the seismic waveform by calculating the envelope of the signal;

[0078] The classification and recognition module analyzes and recognizes the extracted features, and standardizes the extracted features so that features of different dimensions have the same influence in classification. Common standardization methods include normalization and standardization. The classification algorithm is used to train the model, and the data set is divided into a training set and a test set. The model is trained based on the selected classification algorithm using the training set data. The performance of the model is evaluated through k-fold cross-validation to prevent overfitting, and the F1 value of the model is calculated. The ROC curve is plotted and the AUC value is calculated. The performance of the classifier at different thresholds is evaluated, and the model availability threshold of the F1 value and the AUC value of the ROC curve is preset. Only when the F1 value of the model and the AUC value of the ROC curve simultaneously meet the model availability threshold, the model enters the actual application stage. The trained model is used to classify real-time seismic signals and identify and judge potential non-seismic event signals in continuous deep-source seismic signals.

[0079] It should be noted that the optional classification algorithms include support vector machine, decision tree, random forest, k nearest neighbor, naive Bayes, neural network, gradient boosting machine and other algorithms. The calculation method of F1 value is In the formula, P is the precision rate, R is the recall rate, and the calculation method of precision rate is The calculation method of recall is Among them, TP is the true positive example, which indicates the number of samples correctly predicted as positive, FP is the false positive example, which indicates the number of samples incorrectly predicted as positive, and FN is the false negative example, which indicates the number of samples incorrectly predicted as negative.

[0080] The ROC curve is a graphical tool used to evaluate the performance of a binary classification model. It evaluates the model's discriminative ability by showing the performance of the classifier under different thresholds. The ROC curve is composed of the recall rate and the false positive rate. The false positive rate indicates the proportion of samples that are actually negative that are incorrectly predicted as positive. The calculation method is: Where TN is the true negative, which indicates the number of samples that are actually negative but are correctly predicted as negative. In the ROC curve, the horizontal axis represents the false positive rate, and the vertical axis represents the true positive rate. AUC is the numerical representation of the area under the ROC curve, ranging from 0 to 1. The larger the AUC value, the better the classifier performance.

[0081] During the continuous signal process of deep-source earthquakes, the seismic signal and the non-seismic event signal can be separated and identified by analyzing the characteristics of the low-frequency signal and combining it with the trigger algorithm. This can effectively extract valid data that can be used for earthquake early warning and earthquake research, avoid the failure of seismic signal recognition caused by the stacking of low-frequency signals, accurately pick up the fluctuation changes of the seismic waveform, and further improve the accuracy and timeliness of seismic signal recognition.

[0082] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0083] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of computer program goods. The computer program goods include one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0084] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0085] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0086] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0087] If the functions are implemented in the form of software functional units and sold or used as independent goods, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the prior art or the part of the technical solution can be embodied in the form of software goods, and the computer software goods are stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0088] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. The automatic identification system of seismic wave signals based on machine learning is characterized by: It includes preliminary filtering module, spectrum analysis module, seismic phase picking module, feature analysis module and classification and recognition module; The preliminary filtering module is used to preliminarily process and filter the seismic wave signal, process the seismic signal in segments, and calculate the signal energy; The spectrum analysis module is used to analyze the frequency characteristics of the signal, perform frequency analysis on the segmented window of the seismic signal, and extract the spectrum characteristics of the seismic signal through the Fourier transform method; The seismic phase picking module is used to calculate the low-frequency background interference degree of the seismic signal, and combined with the trigger algorithm, distinguish and judge the seismic signal and the non-seismic event signal, and eliminate the non-seismic event signal; The low-frequency background interference degree of the seismic signal is evaluated by calculating the low-frequency background complexity of the signal, and the low-frequency background frequency critical value is set. When the signal frequency is lower than the low-frequency background frequency critical value, the low-frequency background complexity is calculated using the signal window and the background window. The calculation expression is: , where SN is the low-frequency background complexity, is the waveform data of the signal window, is the waveform data of the background window, is the number of data points in the signal window, is the waveform value at the i-th time point in the signal window, is the number of data points in the background window, is the waveform value at the i-th time point in the background window; By setting the trigger threshold Th, potential seismic signals are identified and determined. The calculation expression of the trigger value Tr is: ,and is the short-term average value at time t, is the long-term average value at time t. If the trigger value Tr is greater than or equal to the trigger threshold Th, the seismic signal recognition reliability is high and the characteristics of the seismic wave are more obvious; If the trigger value Tr is less than the trigger threshold Th, the reliability of seismic signal recognition is low and the characteristics of seismic waves are less obvious; The feature analysis module is used to extract the features of seismic waveforms from the original seismic records and transmit them to the classification and recognition module; The classification and recognition module is used to analyze and identify the extracted features, use the classification algorithm to build a model, and classify and judge the seismic signals.

2. The automatic seismic wave signal recognition system based on machine learning according to claim 1 is characterized in that: The method of segmenting seismic signals is as follows: For the formula of the sum of squares of signal energy, for a given time window , whose energy is expressed as , where E is the energy signal intensity, that is, the total energy accumulation of the signal in a certain period of time, is the amplitude value of the signal at time t, indicating the instantaneous value of the signal at a certain moment, t is the sampling moment of the signal, t varies from 1 to N, representing all sampling points in the time window, N is the total number of sampling points in the time window, indicating the total number of sampling points contained in the selected time window length, and N is greater than or equal to 1; The root mean square formula of the signal energy is: , where RMS is the root mean square value, which is the square root of the energy of the signal in the selected time window, and represents the effective value of the signal in AC signal analysis.

3. The automatic seismic wave signal recognition system based on machine learning according to claim 1 is characterized in that: The method of combining the low-frequency background interference level of the seismic signal with the trigger algorithm is: Calculate the short-time average energy of the signal, the expression is , where is the amplitude value of the signal at time point i, is the short-time average at time t, which represents the average energy of the signal in a shorter time window. is the length of the short-time average window, that is, the number of sample points in the time window used to calculate the STA value; Calculate the long-term average energy of the signal, the expression is , where is the long-term average value at time t, which represents the average energy of the signal in a longer time window. is the length of the long-term average window, that is, the number of sample points in the time window used to calculate the LTA value.

4. The automatic seismic wave signal recognition system based on machine learning according to claim 1 is characterized in that: The method for distinguishing earthquake signals from non-earthquake event signals and eliminating non-earthquake event signals is as follows: When the trigger value Tr is greater than or equal to the trigger threshold Th, the signal is highly reliable as a seismic signal, the signal is marked as a valid signal, and the valid signal waveform is stored for analysis; When the trigger value Tr is less than the trigger threshold Th, the signal is a seismic signal with low credibility, and the signal is marked as a background interference signal, which is then screened out.

5. The automatic seismic wave signal recognition system based on machine learning according to claim 1 is characterized in that: The features extracted from the original seismic records to identify and classify seismic waveforms include time domain features, frequency domain features, wavelet transform features, time-frequency domain joint features, high-order statistical features, and morphological features.

6. The automatic seismic wave signal recognition system based on machine learning according to claim 5 is characterized in that: The extracted features are analyzed and identified, and the classification algorithm is used to establish a model. The method for classifying and judging seismic signals is as follows: The extracted features are standardized, and the classification algorithm is used to train the model. The data set is divided into a training set and a test set. The training set data is used to train the model based on the selected classification algorithm. The performance of the model is evaluated through k-fold cross validation to prevent overfitting. The F1 value of the model is calculated, the ROC curve is plotted and the AUC value is calculated. The model usable thresholds of the F1 value and the AUC value of the ROC curve are preset. Only when the F1 value of the model and the AUC value of the ROC curve simultaneously meet the model usable threshold, the model enters the actual application stage, and the trained model is used to classify real-time seismic signals and identify and judge potential non-seismic event signals in continuous deep-source seismic signals.

Images