A method for classifying and identifying seismic signals based on multi-feature fusion

Abstract

The present application relates to the technical field of earthquake event monitoring, in particular to a seismic signal classification and identification method based on multi-feature fusion. The present application first pre-processes each seismic data to obtain pre-processed seismic data, then extracts time domain features and frequency domain features based on the pre-processed seismic data, sets ladder weights of the time domain features and ladder weights of the frequency domain features respectively, combines the ladder weights to fuse the time domain features and the frequency domain features, obtains fused features, inputs the fused features into a seismic signal classification model to train the seismic signal classification model, obtains a trained seismic signal classification model, and finally inputs real-time seismic data into the trained seismic signal classification model to obtain the seismic type of the real-time seismic data. The present application can distribute weights according to the importance of features by using a weighted fusion method, so that the key features have a greater impact on the classification results of the seismic signal classification model, thereby improving the classification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of earthquake event monitoring, and in particular to a method for earthquake signal classification and recognition based on multi-feature fusion. Background Art

[0002] Earthquakes are a natural phenomenon caused by the rapid release of energy in the Earth's crust, which produces seismic waves. Earthquakes are primarily caused by the collision and compression of Earth's plates, which causes shifts and ruptures along their edges and within their interiors. The location where an earthquake begins is called the hypocenter, and the ground directly above it is called the epicenter. The area where the ground shakes most strongly during a destructive earthquake is called the mesembrance zone, which is often the same area as the epicenter.

[0003] A Chinese patent with publication number CN116660992B discloses a seismic signal processing method based on multi-feature fusion. By inputting a vector token with a weight into a multi-head attention mechanism network, the classification result of the seismic signal is finally obtained through the classification token in the module. However, the prior art does not explain how to set weights for the features of different types of seismic signals, resulting in the inability to determine the weights of different features during the actual processing process, making the reliability of the final output classification result insufficient. Summary of the Invention

[0004] The purpose of the present invention is to address the problems existing in the background technology and propose a seismic signal classification and recognition method based on multi-feature fusion.

[0005] The technical solution of the present invention:

[0006] A seismic signal classification and recognition method based on multi-feature fusion is provided, including:

[0007] Acquire multiple seismic data, preprocess each seismic data, and obtain preprocessed seismic data;

[0008] Extracting time domain features and frequency domain features based on preprocessed seismic data, setting step weights for time domain features and frequency domain features respectively, and fusing the time domain features and frequency domain features with the step weights to obtain fused features;

[0009] Create a seismic signal classification model;

[0010] Inputting the fusion features into a seismic signal classification model to train the seismic signal classification model, thereby obtaining a trained seismic signal classification model; the trained seismic signal classification model has the ability to output the earthquake type corresponding to the input seismic data;

[0011] Real-time seismic data is collected and input into a trained seismic signal classification model to obtain the earthquake type of the real-time seismic data.

[0012] Preferably, obtaining a plurality of seismic data and preprocessing each seismic data to obtain preprocessed seismic data includes:

[0013] Create earthquake data table;

[0014] Get multiple earthquake data and put all earthquake data into earthquake data table;

[0015] Selecting a seismic data, performing filtering processing on the seismic data, and obtaining filtered seismic data;

[0016] De-noising the filtered seismic data to obtain pre-processed seismic data, and putting the pre-processed seismic data into a seismic data table;

[0017] Return to select one seismic data until all seismic data are selected, and obtain multiple pre-processed seismic data.

[0018] Preferably, time domain features and frequency domain features are extracted based on the preprocessed seismic data, step weights of the time domain features and the step weights of the frequency domain features are set respectively, and the time domain features and the frequency domain features are fused in combination with the step weights to obtain fused features, including:

[0019] Select a pre-processed earthquake data from the earthquake data table;

[0020] respectively extracting the time domain features and frequency domain features of the preprocessed seismic data;

[0021] Set corresponding step weights for each time domain feature and each frequency domain feature;

[0022] Combine the step weights of time domain features and the step weights of frequency domain features to perform feature fusion, obtain fused features, and put the fused features into the seismic data table;

[0023] Return and select a preprocessed seismic data from the seismic data table until all the preprocessed seismic data in the seismic data table are selected, and obtain the fusion features corresponding to each preprocessed seismic data.

[0024] Preferably, respectively extracting the time domain features and frequency domain features of the preprocessed seismic data includes:

[0025] Obtaining the duration of the preprocessed seismic data and each peak value within the duration;

[0026] The mean value of the pre-processed seismic data within the duration is extracted by formula 1;

[0027]

[0028] Among them, μ is the mean value of the preprocessed seismic data within the duration, N is the total number of signals of the preprocessed seismic data within the duration, and x i is the i-th signal value of the preprocessed seismic data within the duration;

[0029] Performing Fourier transform on the preprocessed seismic data, thereby converting the preprocessed seismic data into a frequency domain signal;

[0030] The frequency center of gravity of the frequency domain signal is extracted by formula 2;

[0031]

[0032] Among them, F is the frequency center of the frequency domain signal, t(k) is the spectrum of the frequency domain signal, and f k is the frequency value of the kth spectral line, K is the total number of spectral lines, and k∈[1,K].

[0033] Preferably, the step weights of the time domain features and the step weights of the frequency domain features are combined to perform feature fusion to obtain fused features, and the fused features are put into the seismic data table, including:

[0034] Combine the step weights and calculate the fusion features using Formula 3;

[0035]

[0036] Among them, P mix is the fusion feature, μ m is the mth time domain feature, F m is the mth frequency domain feature, Z 1m is the step weight corresponding to the mth time domain feature, Z 2m is the step weight corresponding to the mth frequency domain feature, and M is the total number of frequency domain features and time domain features selected for the fusion feature.

[0037] Preferably, before combining the step weights of the time domain features and the step weights of the frequency domain features to perform feature fusion to obtain fused features and putting the fused features into the seismic data table, the following steps are included:

[0038] Obtain multiple historical earthquake data;

[0039] Extracting the time domain characteristics and frequency domain characteristics of each historical earthquake data, and constructing a coupling relationship of the historical earthquake data, the time domain characteristics of the historical earthquake data, and the frequency domain characteristics of the historical earthquake data for each historical earthquake data; taking a coupling relationship of the historical earthquake data, the time domain characteristics of the historical earthquake data, and the frequency domain characteristics of the historical earthquake data as a data sample;

[0040] Create a stepped weight model;

[0041] The data samples corresponding to multiple historical earthquake data are divided into training sets and test sets according to random proportions;

[0042] Inputting data samples in the training set into the step weight model to train the step weight model, thereby obtaining a trained step weight model; the trained step weight model has the ability to automatically output step weights corresponding to different time domain features of the input seismic data and step weights corresponding to different frequency domain features of the seismic data according to the input seismic data;

[0043] The test set is input into the trained step weight model to verify whether the trained step weight model is qualified.

[0044] Preferably, the data samples in the training set are input into the ladder weight model to train the ladder weight model, and the trained ladder weight model is obtained, including:

[0045] Set multiple earthquake levels and the range threshold corresponding to each earthquake level;

[0046] All training samples are divided into multiple groups of training samples according to the range threshold corresponding to the earthquake classification;

[0047] For each group of training samples, all training samples in the group are sorted from large to small according to the earthquake magnitude, and the first N training samples in each group of training samples are selected; the first N training samples in each group are recorded as a group of denoised samples, and multiple groups of denoised samples are obtained;

[0048] Select a group of denoised samples and obtain all time domain features and all frequency domain features of each denoised sample in the group;

[0049] Calculate the occurrence frequency of each time domain feature and the occurrence frequency of each frequency domain feature;

[0050] Obtain the top X time domain features and the top Y frequency domain features; record the selected features as important features;

[0051] Set the step weights and the value range of each step weight;

[0052] Assign corresponding step weights to each important feature based on the value range of each step weight;

[0053] Return to select a group of denoised samples until each group of denoised samples is selected, and obtain the step weights corresponding to all important features of each group of denoised samples.

[0054] Preferably, the fusion features are input into a seismic signal classification model to train the seismic signal classification model, and a trained seismic signal classification model is obtained, comprising:

[0055] Acquire a plurality of preprocessed seismic data, a fusion feature corresponding to each preprocessed seismic data, and a seismic level of each preprocessed seismic data; and use the plurality of preprocessed seismic data, the fusion feature corresponding to each preprocessed seismic data, and the seismic level of each preprocessed seismic data as a data sample;

[0056] Divide multiple data samples into training sets and test sets in random proportions;

[0057] Inputting the training set into the seismic signal classification model to obtain a trained seismic signal classification model;

[0058] The test set is input into the trained seismic signal classification model to verify whether the trained seismic signal classification model is trained.

[0059] Preferably, collecting real-time seismic data and inputting the real-time seismic data into a trained seismic signal classification model to obtain the earthquake type of the real-time seismic data includes:

[0060] Set warning thresholds;

[0061] Acquire real-time seismic data and input the real-time seismic data into the trained seismic signal classification model;

[0062] Obtaining the real-time earthquake classification output by the trained earthquake signal classification model, and determining whether the real-time earthquake classification output by the trained earthquake signal classification model is greater than or equal to the warning threshold;

[0063] If the real-time earthquake grade output by the trained earthquake signal classification model is greater than or equal to the warning threshold, an early warning is issued.

[0064] Preferably, the seismic signal classification and recognition method based on multi-feature fusion further includes:

[0065] Obtain the prediction results output by the trained seismic signal classification model;

[0066] The prediction results and the real-time earthquake data corresponding to the prediction results are added to the training set.

[0067] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0068] First, each seismic data is preprocessed to obtain preprocessed seismic data. Then, time domain features and frequency domain features are extracted based on the preprocessed seismic data, and the step weights of the time domain features and the step weights of the frequency domain features are set respectively. The time domain features and the frequency domain features are fused in combination with the step weights to obtain fused features, and the fused features are input into the seismic signal classification model to train the seismic signal classification model to obtain the trained seismic signal classification model. Finally, the real-time seismic data is input into the trained seismic signal classification model to obtain the earthquake type of the real-time seismic data. The present application uses a weighted fusion method to assign weights according to the importance of the features, so that the key features have a greater impact on the classification results of the seismic signal classification model, thereby improving the accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a flow chart of a seismic signal classification and recognition method based on multi-feature fusion proposed by the present invention; DETAILED DESCRIPTION

[0070] like Figure 1 As shown, the present invention proposes a seismic signal classification and recognition method based on multi-feature fusion, comprising:

[0071] S100, acquiring a plurality of seismic data, and preprocessing each seismic data to obtain preprocessed seismic data;

[0072] S200, extracting time domain features and frequency domain features based on the preprocessed seismic data, setting step weights for the time domain features and the frequency domain features, respectively, and fusing the time domain features and the frequency domain features in combination with the step weights to obtain fused features;

[0073] S300, creating a seismic signal classification model;

[0074] S400, inputting the fusion features into a seismic signal classification model to train the seismic signal classification model, thereby obtaining a trained seismic signal classification model; the trained seismic signal classification model has the ability to output the earthquake type corresponding to the input seismic data;

[0075] S500 , collecting real-time seismic data, and inputting the real-time seismic data into a trained seismic signal classification model to obtain the earthquake type of the real-time seismic data.

[0076] In the present invention, each seismic data is first preprocessed to obtain preprocessed seismic data, and then time domain features and frequency domain features are extracted based on the preprocessed seismic data, and the step weights of the time domain features and the step weights of the frequency domain features are set respectively. The time domain features and the frequency domain features are fused in combination with the step weights to obtain fused features, and the fused features are input into the seismic signal classification model to train the seismic signal classification model to obtain the trained seismic signal classification model. Finally, the real-time seismic data is input into the trained seismic signal classification model to obtain the earthquake type of the real-time seismic data. The present application uses a weighted fusion method to assign weights according to the importance of the features, so that the key features have a greater impact on the classification results of the seismic signal classification model, thereby improving the accuracy of classification.

[0077] In an optional embodiment, the S100 includes:

[0078] S110, creating an earthquake data table;

[0079] S120, acquiring multiple earthquake data and putting all earthquake data into an earthquake data table;

[0080] S130, selecting a piece of seismic data, filtering the seismic data to obtain filtered seismic data;

[0081] Specifically, filtering is the operation of removing specific frequency bands from a signal, which is an important measure to suppress and prevent interference;

[0082] S140, performing denoising processing on the filtered seismic data to obtain pre-processed seismic data, and putting the pre-processed seismic data into a seismic data table;

[0083] Specifically, the noise can be random noise or white noise with uniform frequency distribution, or it can be frequency-dependent noise introduced by device mechanisms or signal processing algorithms;

[0084] Increasing signals in seismic data is particularly important for seismic imaging, inversion and interpretation. Useful signals that are erased by ambient random noise are easily ignored, which may lead to false discontinuity seismic events and artifacts in the final migration image. Therefore, the useful signal is enhanced by attenuating random noise while preserving the edge characteristics of the seismic section.

[0085] S150, returning to select a seismic data until all seismic data are selected, and obtaining a plurality of pre-processed seismic data.

[0086] It should be noted that since the collected seismic data contains a lot of useless data, if the invalid data in the seismic data is not processed, it will affect the accuracy of subsequent steps. Therefore, in this embodiment, the collected seismic data is filtered and denoised to remove the invalid data contained in the seismic data to improve the accuracy of the prediction.

[0087] In an optional embodiment, the step S200 includes:

[0088] S210, selecting a pre-processed earthquake data from the earthquake data table;

[0089] S220, extracting time domain features and frequency domain features of the preprocessed seismic data respectively;

[0090] Specifically, the time domain describes the relationship between mathematical functions or physical signals and time. For example, the time domain waveform of a signal can express how the signal changes over time; the frequency domain is a coordinate system used to describe the characteristics of a signal in terms of frequency;

[0091] In this embodiment, the time domain features include mean, standard deviation, peak value and kurtosis, and the frequency domain features include centroid frequency, frequency variance and standard deviation. Each time domain feature or each frequency domain feature has its own unique corresponding ladder weight.

[0092] S230, setting corresponding step weights for each time domain feature and each frequency domain feature;

[0093] S240, combining the step weights of the time domain features and the step weights of the frequency domain features to perform feature fusion to obtain fused features, and putting the fused features into the seismic data table;

[0094] S250, returning to select a pre-processed seismic data from the seismic data table until all the pre-processed seismic data in the seismic data table are selected, and obtaining the fusion feature corresponding to each pre-processed seismic data.

[0095] It should be noted that time domain features and frequency domain features are extracted from the preprocessed seismic data. Since different features have different importance for seismic signal classification, it is necessary to set corresponding step weights for different features. The more important the feature, the larger the corresponding step weight. This allows the more important features to provide more effective information in the process of seismic signal classification, while reducing the influence of the less important features on the classification results in seismic signal classification.

[0096] In an optional embodiment, the S210 includes:

[0097] S211, obtaining the duration of the pre-processed seismic data and each peak value within the duration;

[0098] S212, extracting the mean value of the pre-processed seismic data within the duration using Formula 1;

[0099]

[0100] Among them, μ is the mean value of the preprocessed seismic data within the duration, N is the total number of signals of the preprocessed seismic data within the duration, and x i is the i-th signal value of the preprocessed seismic data within the duration;

[0101] Specifically, in addition to the mean, the time domain features also include standard deviation, peak value, and kurtosis, and these time domain features are all used as sub-features of the fusion feature;

[0102] The standard deviation of the preprocessed seismic data is calculated using Formula 4;

[0103]

[0104] Where σ is the standard deviation of the processed seismic data, μ is the mean of the preprocessed seismic data within the duration, N is the total number of signals of the preprocessed seismic data within the duration, and x is the number of signals of the preprocessed seismic data within the duration. i is the i-th signal value of the preprocessed seismic data within the duration;

[0105] The kurtosis of the preprocessed seismic data is calculated using Formula 5;

[0106]

[0107] Among them, Kur is the kurtosis of the preprocessed seismic data, μ is the mean value of the preprocessed seismic data within the duration, N is the total number of signals of the preprocessed seismic data within the duration, x i is the i-th signal value of the preprocessed seismic data within the duration;

[0108] S213, performing Fourier transform on the preprocessed seismic data, thereby converting the preprocessed seismic data into a frequency domain signal;

[0109] Specifically, after the time domain features are extracted, Fourier transform is performed on the pre-processed seismic data, thereby converting the pre-processed seismic data into frequency domain signals, and collecting frequency domain features;

[0110] S214, extracting the frequency center of gravity of the frequency domain signal using Formula 2;

[0111]

[0112] Among them, F is the frequency center of the frequency domain signal, t(k) is the spectrum of the frequency domain signal, and f k is the frequency value of the kth spectral line, K is the total number of spectral lines, and k∈[1,K];

[0113] Specifically, the frequency center of gravity is used to indicate the center of gravity position of the spectrum, so as to describe the abscissa frequency corresponding to the peak value with the larger component in the spectrum.

[0114] It should be noted that by extracting multiple time domain features and multiple frequency domain features of the preprocessed seismic data respectively, and using the extracted features as sub-features of the fusion features, the fusion features are obtained by combining the step weights corresponding to each feature, so that the final fusion features can simultaneously reflect the time domain characteristics and frequency domain characteristics of the seismic data, thereby making the seismic signal classification based on the fusion features more reliable.

[0115] In an optional embodiment, the S240 includes:

[0116] S241, calculate the fusion feature by combining the step weights using Formula 3;

[0117]

[0118] Among them, P mix is the fusion feature, μ x is the xth time domain feature, F y is the yth frequency domain feature, Z x is the step weight corresponding to the x-th time domain feature, Z y is the step weight corresponding to the yth frequency domain feature, M is the total number of frequency domain features and time domain features selected for fusion features, and M = x + y;

[0119] Optionally, after obtaining the fused features, the fused features may be normalized.

[0120] It should be noted that in this embodiment, the fused features are obtained by fusing multiple time domain features and multiple frequency domain features. Since each feature has its own unique corresponding ladder weight, the redundancy and noise characteristics of the final fused features are effectively reduced, and the key features that significantly contribute to the classification results are adopted, thereby reducing the complexity of the seismic signal classification model and reducing the errors caused by noise characteristics.

[0121] This method can effectively improve the accuracy of seismic signal classification and the generalization ability of seismic signal models, and is especially suitable for scenarios with complex backgrounds such as seismic signals.

[0122] In an optional embodiment, before S240, the following steps are included:

[0123] K100, to obtain multiple historical earthquake data;

[0124] K110, extracting the time domain characteristics and frequency domain characteristics of each historical earthquake data, and constructing a coupling relationship of the historical earthquake data, the time domain characteristics of the historical earthquake data, and the frequency domain characteristics of the historical earthquake data for each historical earthquake data; taking a coupling relationship of the historical earthquake data, the time domain characteristics of the historical earthquake data, and the frequency domain characteristics of the historical earthquake data as a data sample;

[0125] K120, create a stepped weight model;

[0126] K130, divides the data samples corresponding to multiple historical earthquake data into training sets and test sets according to random proportions;

[0127] K140, inputting data samples in the training set into the step weight model to train the step weight model, thereby obtaining a trained step weight model; the trained step weight model has the ability to automatically output step weights corresponding to different time domain features of the input seismic data and step weights corresponding to different frequency domain features of the seismic data according to the input seismic data;

[0128] K150, input the test set into the trained step weight model to verify whether the trained step weight model is qualified.

[0129] It should be noted that before setting the step weights corresponding to each feature, a step weight model is first created and trained so that the trained step weight model has the ability to automatically output the step weights corresponding to different time domain features of the seismic data and the step weights corresponding to different frequency domain features of the seismic data according to the input seismic data.

[0130] When training the step weight model, the historical earthquake data collected previously is used. Based on the historical earthquakes, the importance threshold is set or the top N features are selected to screen out the features that are more important for the earthquake signal classification results, and the features with higher importance are given larger step weights.

[0131] In an optional embodiment, the K140 includes:

[0132] K141, set multiple earthquake levels and the range threshold corresponding to each earthquake level;

[0133] Specifically, earthquake classification can be divided into weak earthquake, felt earthquake, moderate earthquake and strong earthquake, and the range thresholds corresponding to each earthquake classification are (0, 3], (3, 4.7], (4.7, 6] and (6, 15];

[0134] K142, divide all training samples into multiple groups of training samples according to the range threshold corresponding to the earthquake classification;

[0135] K143: For each group of training samples, sort all the training samples in the group from large to small according to the earthquake magnitude, and select the first N training samples in each group of training samples; record the first N training samples in each group as a group of denoised samples, and obtain multiple groups of denoised samples;

[0136] K144, select a group of denoised samples and obtain all time domain features and all frequency domain features of each denoised sample in the group;

[0137] K145 calculates the frequency of occurrence of each time domain feature and the frequency of occurrence of each frequency domain feature;

[0138] K146, obtain the top X time domain features and the top Y frequency domain features; record the selected features as important features;

[0139] K147, set the ladder weight and the value range of each ladder weight;

[0140] K148, assigns corresponding step weights to each important feature based on the value range of each step weight;

[0141] K149, return to K144 until each group of denoised samples is selected and the step weights corresponding to all important features of each group of denoised samples are obtained;

[0142] It should be noted that, in this embodiment, earthquakes are classified so that all training samples are divided into multiple groups of training samples. Each level of earthquake classification contains at least one group of training samples. Each group of training samples is then trained separately to find features with higher importance in different earthquake classifications, and assign larger step weights to features with higher importance.

[0143] Through the weighted fusion method, weights can be assigned according to the importance of features, so that key features have a greater impact on the seismic signal classification results, thereby improving the accuracy of classification and helping to optimize the recognition results based on the actual contribution of the features.

[0144] In this embodiment, for step K147, if the set step weights and the assignment ranges of each step weight are 1.0-[8,10], 0.8-[6,8], and 0.6-[4,6] respectively, it means that when a feature appears 10 times in the group of denoised samples, the step weight corresponding to the feature is 1.0.

[0145] Specifically, for weak earthquakes, the time domain feature with the highest occurrence is the mean, and the frequency domain feature with the highest occurrence is the frequency centroid. Therefore, a step weight of 1.0 can be assigned to the two features of "mean" and "frequency centroid". For moderate-strong earthquakes, the time domain feature with the highest occurrence is kurtosis, and the frequency domain feature with the highest occurrence is frequency variance. Therefore, a step weight of 1.0 can be assigned to the two features of "kurtosis" and "frequency variance".

[0146] Optionally, in addition to the "number" criterion for judging the importance of a feature, other criteria such as average value can also be set. The specific steps include:

[0147] K170, select a group of denoised samples and obtain all time domain features and all frequency domain features of each denoised sample in the group;

[0148] K171, select a denoised sample;

[0149] K172, calculating the average value of each time domain feature of the group of denoised samples and the average value of each frequency domain feature of the group of denoised samples;

[0150] K173, set the deviation threshold, based on the average value, select each time domain feature that meets the deviation threshold and each frequency domain feature that meets the deviation threshold based on the average value; each time domain feature that meets the deviation threshold and each frequency domain feature that meets the deviation threshold are recorded as important features.

[0151] By setting the deviation threshold, multiple features that are closest to the average value of the feature are screened out, and the corresponding step weights are set according to the degree of proximity to the average value. The closer the feature is to the average value, the larger its corresponding step weight.

[0152] In an optional embodiment, the step S400 includes:

[0153] S410, obtaining a plurality of preprocessed seismic data, a fusion feature corresponding to each preprocessed seismic data, and a seismic level of each preprocessed seismic data; treating the plurality of preprocessed seismic data, the fusion feature corresponding to each preprocessed seismic data, and the seismic level of each preprocessed seismic data as a data sample;

[0154] S420, dividing the multiple data samples into a training set and a test set in a random ratio;

[0155] Specifically, the proportion of the training set should be larger than that of the test set;

[0156] S430, inputting the training set into the seismic signal classification model to obtain a trained seismic signal classification model;

[0157] S440, inputting the test set into the trained seismic signal classification model to verify whether the trained seismic signal classification model is trained;

[0158] Specifically, when executing S440 , a pass rate threshold may be set, and when the pass rate of the test set is greater than or equal to the pass rate threshold, it can be determined that the training of the trained seismic signal classification model is completed.

[0159] It should be noted that by inputting the fusion features into the seismic signal classification model to train the seismic signal classification model, while preserving the original attributes of each type of features, the correlation relationship between features is established in the seismic signal classification model, thereby improving the seismic signal classification model's ability to recognize seismic signals, helping to capture complex signal patterns and interactions between features, and improving the robustness and accuracy of classification.

[0160] In an optional embodiment, the step S500 includes:

[0161] S510, setting a warning threshold;

[0162] S520, acquiring real-time seismic data and inputting the real-time seismic data into the trained seismic signal classification model;

[0163] S530, obtaining the real-time earthquake grade output by the trained earthquake signal classification model, and determining whether the real-time earthquake grade output by the trained earthquake signal classification model is greater than or equal to a warning threshold;

[0164] S540: If the real-time earthquake grade output by the trained earthquake signal classification model is greater than or equal to the warning threshold, a warning is issued.

[0165] It should be noted that real-time earthquake data is input into the trained earthquake signal classification model to obtain real-time earthquake grades, and an early warning is issued to the outside world when the real-time earthquake grade is greater than or equal to the early warning threshold, so as to remind people to prevent the upcoming earthquake. If the real-time earthquake grade output by the trained earthquake signal classification model is less than the early warning threshold, no early warning is issued.

[0166] In an optional embodiment, the seismic signal classification and recognition method based on multi-feature fusion further includes:

[0167] S600, obtaining the prediction result output by the trained seismic signal classification model;

[0168] S610: Add the prediction result and the real-time earthquake data corresponding to the prediction result to a training set.

[0169] It should be noted that the seismic signal classification model is continuously trained by reusing the prediction results output by the trained seismic signal classification model as training samples, thereby continuously improving the seismic signal classification model and gradually improving the prediction accuracy of the seismic signal classification model.

[0170] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A seismic signal classification and recognition method based on multi-feature fusion, characterized in that: include: Acquire multiple seismic data, preprocess each seismic data, and obtain preprocessed seismic data; Based on the preprocessed seismic data, time domain features and frequency domain features are extracted, and the step weights of the time domain features and the step weights of the frequency domain features are set respectively. The time domain features and the frequency domain features are fused by combining the step weights to obtain fused features, including: Obtain multiple historical earthquake data; Extracting the time domain characteristics and frequency domain characteristics of each historical earthquake data, and constructing a coupling relationship of the historical earthquake data, the time domain characteristics of the historical earthquake data, and the frequency domain characteristics of the historical earthquake data for each historical earthquake data; taking a coupling relationship of the historical earthquake data, the time domain characteristics of the historical earthquake data, and the frequency domain characteristics of the historical earthquake data as a data sample; Create a stepped weight model; The data samples corresponding to multiple historical earthquake data are divided into training sets and test sets according to random proportions; Inputting data samples in the training set into the step weight model to train the step weight model, thereby obtaining a trained step weight model; the trained step weight model has the ability to automatically output step weights corresponding to different time domain features of the input seismic data and step weights corresponding to different frequency domain features of the seismic data according to the input seismic data; Input the test set into the trained step weight model to verify whether the trained step weight model is qualified; The data samples in the training set are input into the ladder weight model to train the ladder weight model, and the trained ladder weight model is obtained, including: Set multiple earthquake levels and the range threshold corresponding to each earthquake level; All training samples are divided into multiple groups of training samples according to the range threshold corresponding to the earthquake classification; For each group of training samples, all training samples in the group are sorted from large to small according to the earthquake magnitude, and the first B training samples in each group are selected; the first B training samples in each group are recorded as a group of denoised samples, and multiple groups of denoised samples are obtained; Select a group of denoised samples and obtain all time domain features and all frequency domain features of each denoised sample in the group; Calculate the occurrence frequency of each time domain feature and the occurrence frequency of each frequency domain feature; Obtain the top X time domain features and the top Y frequency domain features; record the selected features as important features; Set the step weights and the value range of each step weight; Assign corresponding step weights to each important feature based on the value range of each step weight; Return to select a group of denoised samples until each group of denoised samples is selected, and obtain the step weights corresponding to all important features of each group of denoised samples; Create a seismic signal classification model; Inputting the fusion features into a seismic signal classification model to train the seismic signal classification model, thereby obtaining a trained seismic signal classification model; the trained seismic signal classification model has the ability to output the earthquake type corresponding to the input seismic data; Real-time seismic data is collected and input into a trained seismic signal classification model to obtain the earthquake type of the real-time seismic data.

2. A method for seismic signal classification and recognition based on multi-feature fusion according to claim 1, characterized in that: Acquire multiple seismic data, preprocess each seismic data, and obtain preprocessed seismic data, including: Create earthquake data table; Get multiple earthquake data and put all earthquake data into earthquake data table; Selecting a seismic data, performing filtering processing on the seismic data, and obtaining filtered seismic data; De-noising the filtered seismic data to obtain pre-processed seismic data, and putting the pre-processed seismic data into a seismic data table; Return to select one seismic data until all seismic data are selected, and obtain multiple pre-processed seismic data.

3. A method for seismic signal classification and recognition based on multi-feature fusion according to claim 2, characterized in that: Based on the preprocessed seismic data, time domain features and frequency domain features are extracted, and the step weights of the time domain features and the step weights of the frequency domain features are set respectively. The time domain features and the frequency domain features are fused by combining the step weights to obtain fused features, including: Select a pre-processed earthquake data from the earthquake data table; respectively extracting the time domain features and frequency domain features of the preprocessed seismic data; Set corresponding step weights for each time domain feature and each frequency domain feature; Combine the step weights of the time domain features and the step weights of the frequency domain features to perform feature fusion, obtain fused features, and put the fused features into the seismic data table; Return and select a preprocessed seismic data from the seismic data table until all the preprocessed seismic data in the seismic data table are selected, and obtain the fusion feature corresponding to each preprocessed seismic data.

4. A method for seismic signal classification and recognition based on multi-feature fusion according to claim 3, characterized in that: The time domain features and frequency domain features of the pre-processed seismic data are extracted respectively, including: Obtaining the duration of the preprocessed seismic data and each peak value within the duration; The mean value of the pre-processed seismic data within the duration is extracted by formula 1; Formula 1: in, is the mean value of the preprocessed seismic data within the duration, N is the total number of signals of the preprocessed seismic data within the duration, is the i-th signal value of the preprocessed seismic data within the duration; Performing Fourier transform on the preprocessed seismic data, thereby converting the preprocessed seismic data into a frequency domain signal; The frequency center of gravity of the frequency domain signal is extracted by formula 2; Formula 2: in, is the frequency center of gravity of the frequency domain signal, is the spectrum of the frequency domain signal, is the frequency value of the kth spectral line, K is the total number of spectral lines, and .

5. The method for seismic signal classification and recognition based on multi-feature fusion according to claim 4, characterized in that: Combine the step weights of the time domain features and the step weights of the frequency domain features to perform feature fusion to obtain fused features, and put the fused features into the seismic data table, including: Combine the step weights and calculate the fusion features using Formula 3; Formula 3: in, is the fusion feature, is the mth time domain feature, is the mth frequency domain feature, is the step weight corresponding to the mth time domain feature, is the step weight corresponding to the mth frequency domain feature, and M is the total number of frequency domain features and time domain features selected for the fusion feature.

6. A method for seismic signal classification and recognition based on multi-feature fusion according to claim 5, characterized in that: The fusion features are input into the seismic signal classification model to train the seismic signal classification model, and the trained seismic signal classification model is obtained, including: Acquire a plurality of preprocessed seismic data, a fusion feature corresponding to each preprocessed seismic data, and a seismic level of each preprocessed seismic data; and use the plurality of preprocessed seismic data, the fusion feature corresponding to each preprocessed seismic data, and the seismic level of each preprocessed seismic data as a data sample; Divide multiple data samples into training sets and test sets in random proportions; Inputting the training set into the seismic signal classification model to obtain a trained seismic signal classification model; The test set is input into the trained seismic signal classification model to verify whether the trained seismic signal classification model is trained.

7. A method for seismic signal classification and recognition based on multi-feature fusion according to claim 6, characterized in that: Collect real-time seismic data and input the real-time seismic data into the trained seismic signal classification model to obtain the earthquake type of the real-time seismic data, including: Set warning thresholds; Acquire real-time seismic data and input the real-time seismic data into the trained seismic signal classification model; Obtaining the real-time earthquake classification output by the trained earthquake signal classification model, and determining whether the real-time earthquake classification output by the trained earthquake signal classification model is greater than or equal to the warning threshold; If the real-time earthquake grade output by the trained earthquake signal classification model is greater than or equal to the warning threshold, an early warning is issued.

8. The method for seismic signal classification and recognition based on multi-feature fusion according to claim 7, characterized in that: The seismic signal classification and recognition method based on multi-feature fusion also includes: Obtain the prediction results output by the trained seismic signal classification model; The prediction results and the real-time earthquake data corresponding to the prediction results are added to the training set.

Images