CN-122017978-A - Prediction method of seismic source parameters based on physical constraint and geometric adaptation of array

CN122017978ACN 122017978 ACN122017978 ACN 122017978ACN-122017978-A

Abstract

A prediction method for the vibration source parameters based on physical constraint and geometric adaptation of array includes such steps as determining the input, output and initial prediction models, sequentially obtaining the station characteristics and the predicted vibration source parameters, comparing the predicted vibration source parameters with the vibration source true values to obtain regression loss (if there is no vibration source true value), obtaining phase loss according to the predicted vibration source parameters and vibration phase labels, obtaining total loss, training the models, storing optimal models, inputting relevant data in optimal models, obtaining the predicted vibration source parameters, and embedding the physical law of vibration wave propagation in the phase loss as the hard constraint model for optimizing the prediction result.

Inventors

LI CHAWEI
DENG NA
DING WENXIU
JIANG YONG
Lv Pinji
WU GUIJIE
PANG CONG
WU YANXIA
SHEN XUELIN
LUO QI

Assignees

湖北省地震局(中国地震局地震研究所)

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. A prediction method of a seismic source parameter based on physical constraint and geometric adaptation of an array is characterized by comprising the following steps: Firstly, determining continuous waveform data and position information corresponding to a plurality of stations in a data set one by one as input, determining predicted source parameters as output, and setting an initial prediction model; The second step, firstly, extracting the characteristics of the continuous waveform data and the position information, and then obtaining predicted source parameters; Step three, judging whether the data set comprises a vibration phase tag and a vibration source real value, if the data set comprises the vibration phase tag and the vibration source real value, jumping to the step four, and if the data set does not comprise the vibration phase tag and the vibration source real value, jumping to the step five; comparing the predicted source parameter with the true value of the source to obtain regression loss, obtaining real arrival time data according to the seismic phase label, obtaining predicted arrival time data according to the predicted source parameter, obtaining phase loss according to the predicted arrival time data and the real arrival time data, weighting the regression loss and the phase loss to obtain total loss, and jumping to the sixth step; Obtaining a seismic phase tag according to a data set through a seismic phase pickup method, obtaining real arrival time data according to the seismic phase tag, obtaining predicted arrival time data according to predicted seismic source parameters, obtaining phase loss according to the predicted arrival time data and the real arrival time data, taking the phase loss as total loss, and jumping to a sixth step; Step six, firstly, a training process of an initial prediction model is carried out, and then the initial prediction model with the minimum total loss is selected as an optimal model; and seventhly, inputting continuous waveform data and position information corresponding to the multiple stations one by one in the optimal model, and obtaining predicted source parameters from the optimal model.
2. The method for predicting seismic source parameters based on physical constraints and geometric adaptation of an array according to claim 1, wherein in the fourth step and the fifth step, the real arrival time data are real P-wave arrival time data and real S-wave arrival time data, and the predicted arrival time data are predicted P-wave arrival time data and predicted S-wave arrival time data; the calculation steps of the predicted P wave arrival time data and the predicted S wave arrival time data are as follows: firstly, calculating the midjolt distance: ; wherein: Is the middle distance of the shock, As the average radius of the earth, Is the angular distance between the source and the station; And calculating the vibration source distance: ; wherein: as the distance between the vibration sources, Is the depth of the seismic source; then calculating theoretical travel time: ; ; wherein: in order to predict the P-wave arrival time data, In order to predict the S-wave arrival time data, As the distance between the vibration sources, Is the velocity of the P-wave, Is the S-wave velocity.
3. The method for predicting a seismic source parameter based on physical constraints and geometric adaptation of an array according to claim 2, wherein in the fourth step and the fifth step, the obtaining the phase loss according to the predicted time data and the real time data means: Firstly, calculating and predicting the phase loss of the P-wave arrival time data: ; wherein: For the phase loss of the P-wave arrival time data, For a set of all the stations, In order to predict the P-wave arrival time data, As the real P-wave arrival time data, Confidence level for seismic facies pick-up; and calculating the phase loss of the predicted S-wave arrival time data: ; wherein: for the phase loss of the S-wave arrival time data, For a set of all the stations, In order to predict the S-wave arrival time data, As the real S-wave arrival time data, Confidence level for seismic facies pick-up; And then average the phase loss of the predicted P wave arrival time data and the phase loss of the predicted S wave arrival time data to obtain the phase loss: ; In the formula, Is the phase loss.
4. The method for predicting a seismic source parameter based on physical constraints and geometric adaptation of an array as set forth in claim 3, wherein in the fourth step, obtaining the regression loss by comparing the predicted seismic source parameter with the true value of the seismic source means: ; In the formula, In order to return the loss to the original state, In order to predict the source parameters of the seismic source, Is the true value of the seismic source.
5. The method for predicting seismic source parameters based on physical constraints and geometric adaptation of an array as set forth in claim 4, wherein in the fourth step, weighting the regression loss and the phase loss to obtain a total loss means: ; ; ; In the formula, In order to account for the total loss, As a weight for the phase loss, Weights for regression loss; In the fifth step, the phase loss as a total loss means: ; In the formula, Is the total loss.
6. The method for predicting the seismic source parameters based on physical constraint and geometric adaptation of the array is characterized in that in the second step, continuous waveform data and position information are subjected to feature extraction, and the predicted seismic source parameters are obtained by firstly carrying out a feature extraction process of a single station, wherein the feature extraction process of the single station comprises the steps of obtaining local time-frequency features of the station according to the continuous waveform data of the station, obtaining geometric distribution features of the station according to the position information of the station, and then fusing the geometric distribution features of the station with the local time-frequency features to obtain station features of the station; And averaging all the station features, obtaining global event features, and obtaining predicted seismic source parameters according to the global event features.
7. The method for predicting seismic source parameters based on physical constraints and geometric adaptation of an array as defined in claim 6, wherein in the second step, the obtaining local time-frequency characteristics of the station according to the continuous waveform data of the station means obtaining local time-frequency characteristics of the station according to the continuous waveform data of the station by adopting a CNN convolutional neural network; In the second step, the step of obtaining the geometric distribution characteristics of the station according to the position information of the station is to map longitude and latitude coordinates of all stations to a multidimensional space through a linear layer and an activation function to obtain initial position characteristics of all stations, calculate distance information of all stations based on the initial position characteristics of all stations, calculate a distance matrix based on all the distance information, define edge weights of the distance matrix by using Gaussian kernels, generate graph structure data according to the edge weights, and extract the geometric distribution characteristics of all the stations from the graph structure data through a GNN graph neural network.
8. The method for predicting seismic source parameters based on physical constraints and geometric adaptation of an array according to claim 7, wherein in the second step, the calculating distance information of all stations based on initial position features of all stations means: ; ; ; ; In the formula, Is the first Station number The surface distance between the individual stations, As the average radius of the earth, First, the Station number The angular distance between the individual stations, Is the first The latitude (radians) of the individual stations, Is the first The longitude (radians) of the individual stations, Is the first The latitude (radians) of the individual stations, Is the first Longitude (radians) of the individual stations; in the second step, the defining the edge weights of the distance matrix by using the gaussian kernel, and then generating the graph structure data according to the edge weights means that: ; In the formula, In order to make the data of the structure of the graph, As a threshold value of the distance, Is a width parameter; in the second step, the step of fusing the geometric distribution feature of the station with the local time-frequency feature to obtain the station feature of the station means that: ; ; ; In the formula, In order for the station to be characterized, Is a local time-frequency characteristic and is characterized by, As a parameter item of the set of parameters, Is a geometric distribution feature.
9. The method for predicting a seismic source parameter based on physical constraints and geometric adaptation of an array as defined in claim 6, wherein in the third step, obtaining the predicted seismic source parameter according to the global event feature means that a multi-layer perceptron (MLP) is used to perform nonlinear transformation and mapping on the global event feature to obtain high-dimensional information, and then the high-dimensional information is output as the predicted seismic source parameter.
10. The method of claim 1, wherein in the sixth step, the seismic phase pickup method comprises PhaseNet method, STA/LTA method or EQtransform method.

Description

Prediction method of seismic source parameters based on physical constraint and geometric adaptation of array Technical Field The invention relates to a prediction method of a seismic source parameter, belongs to the technical field of seismic monitoring and seismology, and particularly relates to a prediction method of a seismic source parameter based on physical constraint and geometric adaptation of an array. Background The prediction of the source parameters, namely the seismic positioning, is a core task of seismic monitoring and geophysics, and not only can the analysis of the seismic risk and the disaster risk prevention be carried out, but also the research of the seismic mechanism and the internal structure of the earth can be facilitated. The China patent with the application number 202410952406.3 and the application date 2024 and the date 07 discloses a method and a device for predicting earthquake focus parameters based on deep learning, which comprise the steps of obtaining an earthquake event waveform, P wave arrival time and a epicenter distance from a target station, preprocessing the earthquake event waveform according to the P wave arrival time to obtain a T component earthquake event waveform, constructing an auxiliary waveform with the same length as the T component earthquake event waveform, wherein the auxiliary waveform comprises 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 station information of the target station, inputting the T component earthquake event waveform and the auxiliary waveform into a preset earthquake focus parameter prediction model, and outputting the predicted earthquake focus parameters by the earthquake focus parameter prediction model, wherein the design can obtain the earthquake focus parameters, but has the following defects: In the design, the core input is the amplitude, phase and frequency spectrum information of the complete waveform, the accuracy of the waveform data is very dependent, the waveform data with high accuracy is difficult to obtain in microseism monitoring, sparse table network or far field recording, if the accuracy of the waveform data is not high, the prediction accuracy of the seismic source parameters is not high, and therefore the application range of the design is narrow. The disclosure of this background section is only intended to increase the understanding of the general background of the application and should not be taken as an admission or any form of suggestion that this information forms the prior art that is already known to a person of ordinary skill in the art. Disclosure of Invention The invention aims to overcome the defect and problem of narrow application range in the prediction of the seismic source parameters in the prior art, and provides a prediction method of the seismic source parameters based on physical constraint and geometric adaptation of an array, wherein the application range of the prediction of the seismic source parameters is wide. In order to achieve the above object, the technical solution of the present invention is: A method of predicting source parameters based on physical constraints and array geometry adaptation, the method comprising the steps of: Firstly, determining continuous waveform data and position information corresponding to a plurality of stations in a data set one by one as input, determining predicted source parameters as output, and setting an initial prediction model; The second step, firstly, extracting the characteristics of the continuous waveform data and the position information, and then obtaining predicted source parameters; Step three, judging whether the data set comprises a vibration phase tag and a vibration source real value, if the data set comprises the vibration phase tag and the vibration source real value, jumping to the step four, and if the data set does not comprise the vibration phase tag and the vibration source real value, jumping to the step five; comparing the predicted source parameter with the true value of the source to obtain regression loss, obtaining real arrival time data according to the seismic phase label, obtaining predicted arrival time data according to the predicted source parameter, obtaining phase loss according to the predicted arrival time data and the real arrival time data, weighting the regression loss and the phase loss to obtain total loss, and jumping to the sixth step; Obtaining a seismic phase tag according to a data set through a seismic phase pickup method, obtaining real arrival time data according to the seismic phase tag, obtaining predicted arrival time data according to predicted seismic source parameters, obtaining phase loss according to the predicted arrival time data and the real arrival time data, taking the phase loss as total loss, and jumping to a sixth step; Step six, firstly, a trainin