CN118859310A - A method and device for predicting earthquake source parameters based on deep learning - Google Patents

Abstract

The embodiment of the present application provides a method and device for predicting earthquake source parameters based on deep learning, including: obtaining earthquake event waveform, P wave arrival time and epicenter distance from the target station; preprocessing the earthquake event waveform according to the P wave arrival time to obtain the T component earthquake event waveform; constructing an auxiliary waveform with the same length as the T component earthquake event waveform; wherein the auxiliary waveform includes a waveform corresponding to the epicenter distance, a waveform corresponding to the maximum amplitude of the T component earthquake event waveform, and a waveform corresponding to the station information of the target station; the T component earthquake event waveform and the auxiliary waveform are input into a preset earthquake source parameter prediction model, and the earthquake source parameter prediction model outputs the predicted earthquake source parameters. The present application can quickly obtain earthquake source parameters without manual intervention, and improves the prediction accuracy of earthquake source parameters.

Description

A method and device for predicting earthquake source parameters based on deep learning

Technical Field

The embodiments of the present application relate to the field of earthquake observation technology, and in particular to a method and device for predicting earthquake source parameters.

Background Art

The source parameters of an earthquake can provide key information about the source and the properties and stress state of the medium at the source depth, which is of great significance for understanding the source characteristics and causal mechanism of an earthquake and predicting strong ground motions and seismic activity trends. The traditional method of calculating source parameters generally obtains the source spectrum after correcting the geometric attenuation, inelastic attenuation, site response and instrument response of the seismic waveform record, and obtains the source parameters based on the source spectrum analysis. The calculation process is relatively complicated and it is impossible to quickly obtain the source parameters.

Summary of the invention

In view of this, the purpose of the embodiments of the present application is to propose a method and device for predicting earthquake source parameters based on deep learning.

Based on the above purpose, the embodiment of the present application provides a method for predicting earthquake source parameters based on deep learning, including:

Obtain earthquake event waveforms, P-wave arrival times and epicenter distances from target stations;

Preprocessing the earthquake event waveform according to the P wave arrival time to obtain a T component earthquake event waveform;

Constructing an auxiliary waveform having the same length as the T-component seismic event waveform; wherein the auxiliary waveform includes a waveform corresponding to the epicentral distance, a waveform corresponding to the maximum amplitude of the T-component seismic event waveform, and a waveform corresponding to the station information of the target station;

The T-component earthquake event waveform and the auxiliary waveform are input into a preset earthquake source parameter prediction model, and the earthquake source parameter prediction model outputs predicted earthquake source parameters.

Optionally, preprocessing the seismic event waveform according to the P wave arrival time to obtain a T component seismic event waveform includes:

De-averaging the earthquake event waveform to obtain a de-averaged waveform;

Performing a de-linear trend processing on the waveform after removing the mean value to obtain a waveform after removing the linear trend;

Performing filtering on the waveform after de-linearization to obtain a filtered waveform;

Performing rotation processing on the filtered waveform to obtain a rotated waveform;

Based on the rotated waveform, according to the P wave arrival time, the T component earthquake event waveform is intercepted according to a preset time window;

The extracted T-component earthquake event waveform is subjected to waveform normalization processing to obtain a normalized T-component earthquake event waveform.

Optionally, based on the rotated waveform, according to the P wave arrival time, the T component seismic event waveform is intercepted according to a preset time window, including:

Based on the rotated waveform, the waveform between a first time point preset before the P wave arrival time and a second time point preset after the P wave arrival time is intercepted.

Optionally, the extracted T-component seismic event waveform includes noise and S waves that at least reach a preset energy threshold.

Optionally, the station information includes a station identifier; constructing an auxiliary waveform having the same length as the T component seismic event waveform, comprising:

During the waveform normalization process, determining a scaling factor of the waveform;

Taking the logarithm of the scaling factor and converting it into a Gaussian probability distribution curve, obtaining a waveform corresponding to the maximum amplitude; wherein the scaling factor is twice the maximum amplitude;

Converting the epicenter distance into a Gaussian probability distribution curve to obtain a waveform corresponding to the epicenter distance;

The data point corresponding to the station identification is assigned a preset identification value, and a waveform corresponding to the station information is constructed.

Optionally, the source parameters include a zero frequency limit and a corner frequency; and the method further includes:

Based on the zero-frequency limit and the corner frequency, the seismic moment, moment magnitude, focal radius, stress drop and radiation energy are calculated.

Optionally, the predicted seismic source parameters output by the seismic source parameter prediction model include:

The source parameter prediction model outputs the zero frequency limit and corner frequency in the form of a Gaussian probability distribution curve.

Optionally, before acquiring the earthquake event waveform, P wave arrival time and epicenter distance from the target station, the method further includes:

Acquire seismic waveform samples and corresponding source parameter samples; wherein the seismic waveform samples include seismic event waveform samples, P wave arrival time samples and epicenter distance samples; the source parameter samples include zero frequency limit and corner frequency;

Preprocessing the seismic event waveform samples according to the P wave arrival time samples to obtain T component seismic event waveform samples;

Constructing an auxiliary waveform sample having the same length as the T-component seismic event waveform sample;

The zero frequency limit and the corner frequency are respectively converted into the form of Gaussian probability distribution after taking the logarithm, and the source parameter label is constructed;

The T-component earthquake event waveform samples, auxiliary waveform samples and source parameter labels are input into a neural network model for training, and the source parameter prediction model is obtained after training.

Optionally, the method further includes:

Obtain earthquake event waveforms, P-wave arrival times and epicenter distances recorded at each station from multiple stations;

Preprocess the waveform of each earthquake event to obtain the T-component earthquake event waveform corresponding to each station;

Construct auxiliary waveforms with the same length as the waveforms of each T-component earthquake event respectively;

Inputting the waveform of each T component earthquake event and the corresponding auxiliary waveform into the earthquake source parameter prediction model respectively, and the earthquake source parameter prediction model outputs the predicted earthquake source parameters corresponding to each station;

According to the source parameters corresponding to each station, the mean value of the source parameters is calculated.

The present application also provides an earthquake source parameter prediction device, comprising:

An acquisition module is used to obtain earthquake event waveforms, P-wave arrival time and epicenter distance from the target station;

A preprocessing module, used for preprocessing the earthquake event waveform according to the P wave arrival time to obtain a T component earthquake event waveform;

A construction module, used to construct an auxiliary waveform having the same length as the T-component seismic event waveform; wherein the auxiliary waveform includes a waveform corresponding to the epicentral distance, a waveform corresponding to the maximum amplitude of the T-component seismic event waveform, and a waveform corresponding to the station information of the target station;

The prediction module is used to input the T component earthquake event waveform and the auxiliary waveform into a preset source parameter prediction model, and the source parameter prediction model outputs the predicted source parameters.

As can be seen from the above, the method and device for predicting earthquake source parameters based on deep learning provided by the embodiment of the present application obtains the earthquake event waveform, P wave arrival time and epicenter distance from the target station, pre-processes the earthquake event waveform according to the P wave arrival time, obtains the T component earthquake event waveform, and constructs an auxiliary waveform with the same length as the T component earthquake event waveform; the auxiliary waveform includes a waveform corresponding to the epicenter distance, a waveform corresponding to the maximum amplitude of the T component earthquake event waveform, and a waveform corresponding to the station information of the target station, and the T component earthquake event waveform and the auxiliary waveform are input into the preset earthquake source parameter prediction model, and the earthquake source parameter prediction model outputs the predicted earthquake source parameters. The present application can quickly obtain earthquake source parameters without manual intervention, and improves the prediction accuracy of earthquake source parameters.

BRIEF DESCRIPTION OF THE DRAWINGS

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 or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

FIG1 is a schematic diagram of a method flow chart of an embodiment of the present application;

FIG2 is a schematic diagram of the prediction process of the earthquake source parameter prediction model according to an embodiment of the present application;

FIG3 is a schematic diagram of a model training method flow in an embodiment of the present application;

FIG4A is a schematic diagram showing a comparison result between the zero-frequency limit predicted by an embodiment of the present application and the zero-frequency limit calculated by a traditional method;

FIG4B is a schematic diagram showing the comparison results of the corner frequency predicted by the embodiment of the present application and the inflection point frequency calculated by the traditional method;

FIG5 is a block diagram of the device structure of an embodiment of the present application;

FIG. 6 is a block diagram of the structure of an electronic device according to an embodiment of the present application.

DETAILED DESCRIPTION

In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be the usual meanings understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing in front of the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connecting" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

As described in the background technology section, the source parameter calculation method in the related art is generally to obtain the source spectrum based on the inversion of the seismic waveform record, and to obtain the source parameters based on the analysis of the source spectrum. The inversion of the source parameters is usually to first calculate the inelastic attenuation of seismic waves and the site response of the station using methods such as the joint inversion of multiple stations and multiple sources, and then deduct the influence of factors such as the propagation path effect and the site response from the observed spectrum, and then fit it using the theoretical source spectrum model to obtain the source spectrum parameters of the earthquake. The calculation is large and time-consuming. Another method is the empirical Green's function method, which uses a small earthquake with a magnitude difference greater than 1 near the target earthquake as an empirical Green's function. Since the two have approximately the same propagation path to the same station, the propagation path effect and the site response can be eliminated by the spectrum ratio of the target earthquake and the small earthquake to obtain the corner frequency of the earthquake. This method requires finding a suitable earthquake pair, and the number of earthquakes that can be calculated is small.

In view of this, the present application provides an earthquake source parameter prediction method based on deep learning, which preprocesses the acquired earthquake event waveform to obtain a T-component earthquake event waveform, constructs an auxiliary waveform with the same length as the T-component earthquake event waveform, inputs the T-component earthquake event waveform and the auxiliary waveform into a source parameter prediction model, and the source parameter prediction model outputs the predicted zero-frequency limit and corner frequency; wherein, the auxiliary waveform includes key information such as maximum amplitude, epicenter distance and station information, which can greatly improve the accuracy of the model prediction results.

The technical solution of the present application is further described in detail below through specific embodiments.

As shown in FIG1 , an embodiment of the present application provides a method for predicting earthquake source parameters based on deep learning, comprising:

S101: Obtain earthquake event waveform, P wave arrival time and epicenter distance from the target station;

In this embodiment, the seismic observation station (referred to as the station) records all seismic waveforms, among which the seismic waveform containing the seismic event is the seismic event waveform. The epicenter distance can be calculated based on the source location and the station location, and earthquake-related information such as the P-wave arrival time can be picked up from the seismic event waveform using seismic analysis related software.

S102: pre-processing the earthquake event waveform according to the P wave arrival time to obtain the T component earthquake event waveform;

In this embodiment, after obtaining the earthquake event waveform from the target station, the earthquake event waveform is preprocessed to obtain the T component earthquake event waveform. The preprocessing method includes:

De-averaging the earthquake event waveform to obtain a de-averaged waveform;

Performing a de-linear trend processing on the waveform after removing the mean value to obtain a waveform after de-linear trend;

Filtering the waveform after removing the linear trend to obtain a filtered waveform;

Performing rotation processing on the filtered waveform to obtain a rotated waveform;

Based on the rotated waveform, the T-component earthquake event waveform is intercepted according to the preset time window according to the P-wave arrival time;

The extracted T-component earthquake event waveform is subjected to waveform normalization processing according to a preset zoom factor to obtain a normalized T-component earthquake event waveform.

In this embodiment, in order to facilitate better analysis of seismic signals, based on the original seismic event waveform, basic data processing such as removing the mean and removing the linear trend is performed respectively, and filtering is performed after the basic data processing to retain the seismic signal of a specific frequency band; the seismic event waveform is a three-component waveform of E (east-west), N (north-south), and Z (vertical). By rotating to the ZRT coordinate system, the three-component waveform of R (radial), T (tangential), and Z can be obtained. Therefore, the waveform after filtering is rotated to obtain the seismic event waveform of the T component, and then the T component seismic event waveform is intercepted from the rotated waveform according to the set time window according to the P wave arrival time. In order to improve the convergence speed and stability of the neural network model, the intercepted T component seismic event waveform is subjected to waveform normalization processing to obtain the normalized T component seismic event waveform. Since the prediction effect of using the ENZ three-component waveform is the same as the prediction effect of using only the T component waveform, in order to save resources and improve processing efficiency, only the T component seismic event waveform is extracted for prediction.

In some methods, the general method of waveform normalization is to take the maximum and minimum values of all T-component seismic waveforms, and normalize all T-component seismic waveforms as a whole according to the maximum and minimum values, and the processing results are biased. The method for waveform normalization provided in this embodiment is to take the maximum and minimum values of each T-component seismic waveform respectively, and normalize the waveform according to the maximum and minimum values, that is, normalize each waveform separately, which can improve the accuracy of the results.

In some implementations, the waveform normalization processing method is specifically as follows:

Among them, A is the value of any data point in the T-component seismic event waveform, A _max is the maximum value of the data point in the T-component seismic event waveform, A _min is the minimum value of the data point in the T-component seismic event waveform, and A _normal is the value of the normalized data point. After normalization, the maximum value of each waveform is 1 and the minimum value is 0. In the process of waveform normalization, the scaling factor of the waveform is (A _max -A _min ), which is twice the maximum amplitude.

In some methods, the mean removal process refers to removing the average value of the waveform data, and the linear trend removal process refers to fitting the data into a straight line, subtracting the linear trend represented by the straight line from the data, and performing a 0.5-45Hz bandpass filter process on the waveform after basic data processing. Optionally, the mean removal, linear trend removal, filtering, rotation and other processes can be obtained by processing with seismic analysis related software, and this embodiment does not explain in detail the principle and specific method of basic data processing.

In some embodiments, based on the rotated waveform, according to the P wave arrival time, the T component seismic event waveform is intercepted according to a preset time window, including:

Based on the rotated waveform, the waveform from a preset first time point before the arrival time of the P wave to a preset second time point after the arrival time of the P wave is intercepted as the T-component seismic event waveform. The T-component seismic event waveform intercepted through the time window includes noise and S waves that at least reach a preset energy threshold.

In this embodiment, considering that the local noise is superimposed on the seismic waveform recorded by the station, in order to remove the influence of the noise, part of the noise before the arrival of the P wave is retained in the input data, so that the noise can be deducted from the seismic event waveform during the model processing; on the other hand, in order to ensure that the model input data contains enough S wave signals, the intercepted waveform needs to be long enough, and the specific length can be determined according to the epicenter distance of the station used. In some methods, the waveform from 3 seconds before the arrival of the P wave to 40 seconds after the arrival of the P wave is intercepted as the T component seismic event waveform, and the T component seismic event waveform contains 3 seconds of noise and at least 90% of the S wave signal energy.

S103: constructing an auxiliary waveform having the same length as the T-component earthquake event waveform; wherein the auxiliary waveform includes a waveform corresponding to the epicentral distance, a waveform corresponding to the maximum amplitude of the T-component earthquake event waveform, and a waveform corresponding to the station information of the target station;

In this embodiment, after the original earthquake event waveform is preprocessed to obtain the T component earthquake event waveform, an auxiliary waveform having the same length as the T component earthquake event waveform is constructed. Considering that the source parameters of an earthquake are closely related to the amplitude of the earthquake event waveform, and the earthquake waveform recorded by the station contains the influence of the propagation path and the site effect, the epicenter distance and the station information are also important information for calculating the source parameters, therefore, the constructed auxiliary waveform includes the maximum amplitude lost after the waveform is normalized, as well as the epicenter distance and the station information, and inputting the auxiliary waveform into the earthquake source parameter prediction model to predict the earthquake source parameters can greatly improve the accuracy of the prediction results.

In some embodiments, the station information includes a station identifier; constructing an auxiliary waveform having the same length as the T-component seismic event waveform includes:

During the waveform normalization process, the scaling factor of the waveform is determined;

The logarithm of the scaling factor is taken and converted into a Gaussian probability distribution curve to obtain the waveform corresponding to the maximum amplitude;

The epicenter distance is converted into a Gaussian probability distribution curve to obtain the waveform corresponding to the epicenter distance;

The data points corresponding to the station identification are assigned to the preset identification values, and the waveform corresponding to the station information is constructed.

As shown in FIG2, in this embodiment, an auxiliary waveform having the same length as the T-component earthquake event waveform is constructed, and the auxiliary waveform specifically includes three waveforms, one of which is a waveform corresponding to the maximum amplitude of the earthquake event waveform having a first length, the second is a waveform corresponding to the epicentral distance having a second length, and the third is a waveform corresponding to the station information. To construct an auxiliary waveform including three waveforms, for the waveform corresponding to the maximum amplitude, the scaling factor during waveform normalization is converted into a Gaussian probability distribution curve after taking the logarithm as the waveform corresponding to the maximum amplitude, for the waveform corresponding to the epicentral distance, the epicentral distance is converted into a Gaussian probability distribution curve as the waveform corresponding to the epicentral distance, and for the waveform corresponding to the station information, the data point corresponding to the station identification is set to a preset identification value, and other data points are set to other identification values to obtain the waveform corresponding to the station information.

In some embodiments, a seismic event waveform includes a series of data points that are continuous in time. After preprocessing the seismic event waveform, a T-component seismic event waveform consisting of a series of data points is also obtained, and the waveform length of the T-component seismic event waveform is the number of data points. According to the waveform length of the T-component seismic event waveform, an auxiliary waveform with the same data points is constructed, and the auxiliary waveform includes three parts of data points, wherein one part of the data points represents the maximum amplitude, belonging to the waveform corresponding to the maximum amplitude, the second part of the data points represents the epicenter distance, belonging to the waveform corresponding to the epicenter distance, and the third part of the data points represents the station information, belonging to the waveform corresponding to the station information.

For example, for a station with a sampling rate of 100 Hz, a T-component earthquake event waveform consisting of 4300 sampling points can be obtained, and an auxiliary waveform consisting of 4300 data points is constructed, of which 2000 data points are selected to represent the maximum amplitude, 2000 data points represent the epicenter distance, and 300 sampling points represent the station identification. Considering that the amplitude of the waveform will be disturbed by noise, and the earthquake frequency increases exponentially with the decrease of magnitude, the scaling factor of the waveform is expressed in the form of Gaussian probability distribution after taking the logarithm, and the value of the horizontal axis corresponding to the maximum point of the Gaussian probability distribution is the scaling factor after taking the logarithm. For example, the scaling factor of the waveform ranges from 10 ^-4 to 10 ³ , and the range obtained after taking the logarithm with base 10 is between -4 and 3. The first data point represents -4, the Nth data point (1≤N≤2000) represents -4+(N-1)×7/1999, and the 2000th data point represents 3. For the epicentral distance, considering that the size of the epicentral distance will be affected by the error of the source position, it is also expressed in the form of Gaussian probability distribution. For the stations, each station is pre-numbered, the station identification of each station is determined, and each station corresponds to a data point. If a certain earthquake event waveform is recorded by the Kth station, the value of the data point corresponding to the Kth station is set to 1, and the values of the data points corresponding to the other stations are set to 0, that is, the value of the Kth data point among the 300 data points is set to 1, and the values of the other data points are set to 0, and the station information represented by the 300 data points is obtained.

S104: Inputting the T-component earthquake event waveform and the auxiliary waveform into a preset earthquake source parameter prediction model, and the earthquake source parameter prediction model outputs predicted earthquake source parameters.

In this embodiment, after determining the T-component earthquake event waveform and the auxiliary waveform, the T-component earthquake event waveform and the auxiliary waveform are input into a pre-constructed source parameter prediction model, and the source parameter prediction model is used to predict the earthquake source parameters based on the T-component earthquake event waveform and the auxiliary waveform that characterizes the key information of the earthquake.

In some embodiments, the predicted source parameters output by the source parameter prediction model include: the zero-frequency limit and corner frequency in the form of a Gaussian probability distribution curve output by the source parameter prediction model. That is, the data output by the model are the zero-frequency limit and corner frequency in the form of a Gaussian probability distribution. In conjunction with FIG2, specifically, the zero-frequency limit is converted to a Gaussian probability distribution form after taking the logarithm, and the value of the horizontal coordinate corresponding to the point with the largest Gaussian probability distribution is the zero-frequency limit after taking the logarithm as the waveform corresponding to the zero-frequency limit output; the corner frequency is converted to a Gaussian probability distribution form after taking the logarithm, and the value of the horizontal coordinate corresponding to the point with the largest Gaussian probability distribution is the corner frequency after taking the logarithm. Since the frequency of earthquakes increases exponentially with the decrease of magnitude, the distribution of the source parameters output by the model is more balanced after taking the logarithm, which can improve the accuracy of the prediction results.

In some embodiments, the source parameters output by the source parameter prediction model include a zero-frequency limit and a corner frequency. After the model outputs the predicted zero-frequency limit and corner frequency, the source parameters such as the seismic moment, moment magnitude, source radius, stress drop and radiation energy can be further calculated based on the zero-frequency limit and the corner frequency.

In some methods, based on the Brune model, the seismic moment M ₀ , focal radius r, stress drop Δσ, moment magnitude _MW and radiation energy _ER are calculated using formula (2-6) and expressed as:

Where ρ is the rock density, R is the focal distance, _Vs is the S-wave velocity at the focal depth, U _φθ is the radiation pattern factor, f is the frequency, Ω ₀ is the zero-frequency limit, and _fc is the corner frequency.

As shown in FIG3 , in some embodiments, the earthquake source parameter prediction method further includes:

S301: Acquire seismic waveform samples and corresponding source parameter samples; wherein the seismic waveform samples include seismic event waveform samples, P wave arrival time samples and epicenter distance samples; and the source parameter samples include zero frequency limit and inflection frequency;

S302: preprocessing the earthquake event waveform samples according to the P wave arrival time samples to obtain the T component earthquake event waveform samples;

S303: constructing an auxiliary waveform sample with the same length as the T-component earthquake event waveform sample;

S304: Taking the logarithms of the zero-frequency limit and the corner frequency, respectively, and converting them into the form of Gaussian probability distribution, constructing a source parameter label.

S305: Inputting the T-component earthquake event waveform samples, the auxiliary waveform samples and the source parameter labels into the neural network model for training, and obtaining the source parameter prediction model after the training.

This embodiment provides a method for training a source parameter prediction model. First, earthquake event waveform samples, P wave arrival time samples, and epicenter distance samples for training the model are collected, the earthquake event waveform samples are preprocessed to obtain T-component earthquake event waveform samples, and then auxiliary waveform samples with the same length as the T-component earthquake event waveform samples are constructed, and the T-component earthquake event waveform samples and the constructed auxiliary waveform samples are input into a neural network model for training, and a source parameter prediction model is obtained after training.

As shown in Figure 2, this embodiment is trained based on a fully convolutional neural network model, the structure of which includes 17 convolutional layers (Cov2D), maximum pooling (Max Pooling), activation function (Leaky Relu), batch normalization (BatchNormalization) and up sampling layer (Up Sampling), etc. The mean square error is selected as the training loss function, and the Adam method is used as the optimizer. After testing different parameter selections, in the final training, the batch size is set to 16, the learning rate is 0.001, and the number of iterations is 50 times.

In some embodiments, for the T-component seismic event waveform and the auxiliary waveform each including 4300 data points, the input H×W×C of the neural network model is 2×4300×1, that is, the height is 2, the width is 4300, and the number of channels is 1. The model first extracts features from the input data by downsampling layer by layer, and gradually compresses the input data from 2×4300×1 to 1×1×128; then, the model converts the extracted features into the source parameters of the predicted output by upsampling layer by layer, and expands the feature data from 1×1×128 to 1×256×2, that is, the zero frequency limit and corner frequency of the model output are represented by two one-dimensional arrays with a length of 256.

In some embodiments, the method for predicting earthquake source parameters further includes:

Obtain earthquake event waveforms, P-wave arrival times and epicenter distances recorded at each station from multiple stations;

Preprocess the waveform of each earthquake event to obtain the T-component earthquake event waveform corresponding to each station;

Construct auxiliary waveforms with the same length as the waveforms of each T-component earthquake event respectively;

Input the waveform of each T component earthquake event and the corresponding auxiliary waveform into the earthquake source parameter prediction model respectively, and the earthquake source parameter prediction model outputs the predicted earthquake source parameters corresponding to each station;

According to the source parameters corresponding to each station, the mean value of the source parameters is calculated.

In this embodiment, considering the differences in earthquake observation capabilities and earthquake magnitudes in various places, the number of station records of different earthquakes will also be different. The prediction of a single station only requires one station to predict the source parameters of the earthquake, and the application is more flexible. When multiple stations are involved, the earthquake event waveform, P wave arrival time and epicenter distance of each station can be obtained from each station respectively, and the T component earthquake event waveform corresponding to each station is obtained after pre-processing respectively, and then the auxiliary waveform corresponding to each T component earthquake event waveform is constructed respectively, and the T component earthquake event waveform and the corresponding auxiliary waveform of each station are input into the source parameter prediction model, and the source parameters of each station are output by the model. Finally, the average value is calculated based on the source parameters of each station, and the source parameter mean is used as the final prediction result.

In the present embodiment, the source parameter prediction model is tested using real observation data to verify the accuracy of the model. 5883 earthquake events in Sichuan and Yunnan and the source parameters calculated by artificial traditional methods are collected as training sets, and the source parameter prediction model is trained to obtain the trained source parameter prediction model; then 736 earthquake events are selected as test sets, and input into the trained source parameter prediction model to obtain the predicted source parameters, and compare them with the source parameters calculated by artificial traditional methods. As shown in Figures 4A-4B, each point represents an earthquake event, and its abscissa corresponds to the source parameters calculated by artificial calculation, and the ordinate corresponds to the source parameters predicted by the model. The closer the source parameters obtained by the two methods are, the closer the position of the point is to the diagonal line. According to the comparison results, the source parameters predicted by the model have good consistency with the source parameters calculated by artificial traditional methods, proving that the method of the present application can effectively extract earthquake source parameters from earthquake observation data.

The embodiment of the present application provides a method for predicting earthquake source parameters based on deep learning. After obtaining the earthquake event waveform of the station, the T component earthquake event waveform is obtained through preprocessing, and an auxiliary waveform with the same length as the T component earthquake event waveform is constructed. The maximum amplitude of the earthquake event waveform, station information, epicenter distance and other earthquake key information are added to the auxiliary waveform. The T component earthquake event waveform and the auxiliary waveform are input into the earthquake source parameter prediction model, and the zero frequency limit and corner frequency predicted by the model output are used. On the one hand, the earthquake key information is added to the auxiliary waveform, which can greatly improve the accuracy of the model prediction results; on the other hand, only the earthquake event waveform recorded by a single station is needed, and the source parameters are obtained by model prediction, so that the source parameters can be quickly obtained (the model prediction time is less than 0.1s), and the processing complexity is greatly reduced compared with the traditional method. At the same time, it is not limited by the number of station records, and the application is more flexible; on the third hand, based on the zero frequency limit and corner frequency predicted by the model, other key source parameters can be obtained by calculation.

It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the described method.

It should be noted that the above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

As shown in FIG5 , the embodiment of the present application further provides an earthquake source parameter prediction device, comprising:

An acquisition module is used to obtain earthquake event waveforms, P-wave arrival time and epicenter distance from the target station;

A preprocessing module is used to preprocess the earthquake event waveform according to the P wave arrival time to obtain the T component earthquake event waveform;

A construction module, used to construct an auxiliary waveform having the same length as the T-component earthquake event waveform; wherein the auxiliary waveform includes a waveform corresponding to the epicenter distance, a waveform corresponding to the maximum amplitude of the T-component earthquake event waveform, and a waveform corresponding to the station information of the target station;

The prediction module is used to input the T component earthquake event waveform and the auxiliary waveform into a preset source parameter prediction model, and the source parameter prediction model outputs the predicted source parameters.

For the convenience of description, the above devices are described in terms of functions divided into various modules. Of course, when implementing the embodiments of the present application, the functions of each module can be implemented in the same or multiple software and/or hardware.

The apparatus of the above-mentioned embodiment is used to implement the corresponding method in the above-mentioned embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

6 shows a more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment, and the device may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input/output interface 1030, and the communication interface 1040 are connected to each other in communication within the device through the bus 1050.

The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.

The input/output interface 1030 is used to connect the input/output module to realize information input and output. The input/output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).

The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input/output interface 1030, and the communication interface 1040).

It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input/output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

The electronic device of the above embodiment is used to implement the corresponding method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples. Based on the concept of the present disclosure, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply/ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform to be implemented in the embodiments of the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present disclosure, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered illustrative rather than restrictive.

Although the present disclosure has been described in conjunction with specific embodiments of the present disclosure, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the present disclosure.

Claims (10)

1. The earthquake focus parameter prediction method based on deep learning is characterized by comprising the following steps of:

acquiring a seismic event waveform, P wave arrival time and a midrange from a target station;

Preprocessing the seismic event waveform according to the P wave arrival time to obtain a T component seismic event waveform;

constructing an auxiliary waveform with the same length as the waveform of the T-component seismic event; the auxiliary waveforms comprise waveforms corresponding to the midjourneys, waveforms corresponding to the maximum amplitudes of the T-component seismic event waveforms and waveforms corresponding to station information of the target station;

and inputting the T-component seismic event waveform and the auxiliary waveform into a preset seismic source parameter prediction model, and outputting predicted seismic source parameters by the seismic source parameter prediction model.

2. The method of claim 1, wherein preprocessing the seismic event waveform based on the P-wave arrival time to obtain a T-component seismic event waveform comprises:

performing mean value removal processing on the seismic event waveform to obtain a waveform after mean value removal;

Carrying out linear trend removal on the waveform after mean removal to obtain a waveform after linear trend removal;

filtering the waveform with the linear trend to obtain a filtered waveform;

performing rotation processing on the filtered waveform to obtain a rotated waveform;

based on the rotated waveform, cutting out a T-component seismic event waveform according to a preset time window according to the arrival time of the P wave;

And carrying out waveform normalization processing on the cut T-component seismic event waveform to obtain a normalized T-component seismic event waveform.

3. The method of claim 2, wherein based on the rotated waveform, truncating the T-component seismic event waveform according to a preset time window from the P-wave arrival time, comprising:

Based on the rotated waveform, waveform from a first time point preset before the arrival time of the P wave to a second time point preset after the arrival time of the P wave is intercepted.

4. A method according to claim 3, wherein the truncated T-component seismic event waveform comprises noise and S-waves at least up to a preset energy threshold.

5. The method of claim 2, wherein the station information comprises a station identification; constructing an auxiliary waveform having the same length as the T-component seismic event waveform, comprising:

In the waveform normalization processing process, determining scaling factors of waveforms;

Taking the logarithm of the scaling multiple, and converting the logarithm of the scaling multiple into a Gaussian probability distribution curve to obtain a waveform corresponding to the maximum amplitude; wherein the scaling factor is 2 times the maximum amplitude;

converting the epicenter distance into a Gaussian probability distribution curve to obtain a waveform corresponding to the epicenter distance;

and assigning the data point corresponding to the station identifier as a preset identifier value, and constructing a waveform corresponding to the station information.

6. The method of claim 1, wherein the source parameters include zero frequency limits and corner frequencies; the method further comprises the steps of:

And calculating the earthquake moment, moment magnitude, earthquake focus radius, stress drop and radiation energy according to the zero frequency limit and the corner frequency.

7. The method of claim 6, wherein outputting predicted source parameters by the source parameter prediction model comprises:

Zero frequency limits and corner frequencies in the form of gaussian probability distribution curves are output by the source parameter prediction model.

8. The method of claim 1, wherein prior to acquiring the seismic event waveform, the P-wave arrival time, and the epicenter distance from the target station, further comprising:

acquiring a seismic waveform sample and a corresponding seismic source parameter sample; the seismic waveform samples comprise a seismic event waveform sample, a P wave arrival time sample and a middle-of-earthquake distance sample; the source parameter samples include zero frequency limits and corner frequencies;

Preprocessing the seismic event waveform sample according to the P wave arrival time sample to obtain a T component seismic event waveform sample;

constructing auxiliary waveform samples with the same length as the T-component seismic event waveform samples;

the zero frequency limit and the corner frequency are respectively converted into Gaussian probability distribution after logarithm is taken, and a seismic source parameter label is constructed;

And inputting the T-component seismic event waveform sample, the auxiliary waveform sample and the seismic source parameter label into a neural network model for training, and obtaining the seismic source parameter prediction model after training.

9. The method as recited in claim 1, further comprising:

Acquiring seismic event waveforms, P wave arrival time and epicenter distance recorded by each station from a plurality of stations;

Preprocessing each seismic event waveform to obtain a T-component seismic event waveform corresponding to each station;

Respectively constructing auxiliary waveforms with the same length as the waveform of each T-component seismic event;

respectively inputting each T component seismic event waveform and corresponding auxiliary waveform into the seismic source parameter prediction model, and outputting predicted seismic source parameters corresponding to each station by the seismic source parameter prediction model;

And calculating a seismic source parameter mean value according to the seismic source parameters corresponding to each station.

10. A depth learning based seismic source parameter prediction apparatus, comprising:

The acquisition module is used for acquiring the seismic event waveform, the P wave arrival time and the epicenter distance from the target station;

The preprocessing module is used for preprocessing the seismic event waveform according to the P wave arrival time to obtain a T component seismic event waveform;

the construction module is used for constructing an auxiliary waveform with the same length as the waveform of the T-component seismic event; the auxiliary waveforms comprise waveforms corresponding to the midjourneys, waveforms corresponding to the maximum amplitudes of the T-component seismic event waveforms and waveforms corresponding to station information of the target station;

and the prediction module is used for inputting the T-component seismic event waveform and the auxiliary waveform into a preset seismic source parameter prediction model, and outputting predicted seismic source parameters by the seismic source parameter prediction model.