CN115438716A - Training data generation method for EEW real-time p-wave acquisition - Google Patents
Training data generation method for EEW real-time p-wave acquisition Download PDFInfo
- Publication number
- CN115438716A CN115438716A CN202210938153.5A CN202210938153A CN115438716A CN 115438716 A CN115438716 A CN 115438716A CN 202210938153 A CN202210938153 A CN 202210938153A CN 115438716 A CN115438716 A CN 115438716A
- Authority
- CN
- China
- Prior art keywords
- time
- data
- wave
- real
- eew
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a training data generation method for EEW real-time p-wave acquisition, which comprises the following steps: step one, screening a sample; step two, estimating arrival time; step three, resist and expand; step four, training and constructing; in the third step, the original data is taken as a discrimination set and input into a discriminator for generating a confrontation neural network model, then noise data generated randomly is taken as a generation set and input into a generator for generating the confrontation neural network model, and the extremely-small game confrontation training is carried out to generate a large amount of expansion data which is very close to the discrimination set; according to the invention, p-wave in-time picking can be carried out only by one empirical parameter of window time, the wrong setting risk is small, the problems of wrong picking and missed picking of wave in-time picking are avoided, and the picking is more reliable; the initial arrival time of the p-wave signal is picked up twice by using a long-time and short-time average value ratio method and an iterative accumulation square sum algorithm, so that the calculation speed is high, the estimation error is small, the picking efficiency is high, and the result is more accurate.
Description
Technical Field
The invention relates to the technical field of data generation methods, in particular to a training data generation method for EEW real-time p-wave acquisition.
Background
The Earthquake Early Warning (EEW) means that after a destructive earthquake occurs, electromagnetic waves, earthquake longitudinal waves (p waves) and earthquake transverse waves (s waves) can be utilized to have different propagation speeds in certain areas, and danger avoiding warning information is sent out before the earthquake arrives so as to reduce disaster loss of related early warning areas.
In modern EEW systems for p-wave arrival time picking, many methods have been proposed over the past decades, including short term mean of signal feature function/long term mean of signal feature function (STA/LTA) (Allen, 1978, 1982), akaike Information Criterion (AIC) (Maeda, 1985), vertical and horizontal amplitude ratio (V/H) (Nakamura, 1988), autoregressive method (AR) (Sleeman and Eck, 1999), traditional neural networks (Gentili and Michelini, 2006), and filters (Lomax et al, 2012). EEW, without any p-wave arrival at a selected time delay, has a higher requirement for accuracy and timeliness than that required for near real-time automatic seismic location identification, which typically has a delay of several seconds.
While the above methods have been substantially successful, they still encounter challenges in EEW systems such as false P-wave pickup as noise, missed P-wave pickup and insufficient P-wave pickup accuracy when arriving. Furthermore, these methods are based on signal characteristic variations of p-wave amplitude, energy and/or spectral components (Xiao et al, 2016), which have the common disadvantage that they all require some form of empirical parameters. For example, because of the speed and simplicity of STA and LTA, it is widely used in EEW systems such as china (Zhang et al, 2016 wu and kanamori, 2005) and california (Shake et al, 2014), but this approach requires three empirical parameters: a STA window time, an LTA window time, and a STA/LTA trigger threshold. Any incorrect setting of these three parameters can result in false and/or missing picks, or problems with large errors in the time of day picks (kwonetal, 2018). ShakeAlert attempts to remedy the shortcomings of STA/LTA by using the results of four sites, but there are still many false alarms each year (Cochran et al, 2017), which are disadvantageous in any EEW system.
Disclosure of Invention
The present invention aims to provide a training data generation method for EEW real-time p-wave acquisition, so as to solve the problems in the background art.
In order to achieve the purpose, the invention provides the following technical scheme: a training data generation method for EEW real-time p-wave acquisition comprises the following steps: step one, screening a sample; step two, estimating arrival time; step three, resisting expansion; step four, training and constructing;
in the first step, a p-wave observation signal generated by an earthquake is detected in real time through an earthquake monitoring station, an observation acceleration diagram of the p-wave observation signal is drawn on a data window by taking time as a horizontal axis and acceleration as a vertical axis, and then a large amount of historical data with strong characteristic correlation, namely the p-wave historical signal, is screened out from a database of the earthquake monitoring station by using a Pearson correlation coefficient algorithm and is used as a sample set of a real-time p-wave acquisition model of an EEW system by taking acceleration as data characteristics of the p-wave observation signal;
in the second step, all samples in the sample set are processed by using a long-short time average ratio method, a rough arrival time point of the p-wave historical signal in each sample is picked up, an iterative sum-of-squares algorithm is applied, signals in a specific time length region before and after the rough arrival time point are intercepted, iterative sum-of-squares statistics is calculated, a time point corresponding to the maximum value of the statistics is found out, the time point is an accurate arrival time point of the p-wave historical signal, and then the signals in the specific time length region before and after the accurate arrival time point are intercepted and used as original data of a real-time p-wave acquisition model of the EEW system;
in the third step, the original data is taken as a discrimination set and input into a discriminator for generating a confrontation neural network model, then noise data generated randomly is taken as a generation set and input into a generator for generating the confrontation neural network model, and the extremely-small game confrontation training is carried out to generate a large amount of expansion data which is very close to the discrimination set;
in the fourth step, according to a recording time axis, original data is respectively split into a long-term data set, a medium-term data set and a short-term data set, then the expanded data is split into three data sets which are equally split, and the three data sets are respectively added into the long-term data set, the medium-term data set and the short-term data set to obtain a training data set, a verification data set and a test data set, wherein the training data set is training data of a real-time p-wave acquisition model of the EEW system.
Preferably, in the first step, the pearson correlation coefficient has a value range of [ -1,1], and the closer the absolute value of the correlation coefficient is to 1, the more correlated the two attributes are, the positive and negative values of the value represent the positive or negative correlation of the two attributes, the positive value represents the positive correlation, and the negative value represents the negative correlation.
Preferably, in the second step, the specific time length region selects t 5.12s before the coarse arrival time point 1 T to 3s after the coarse arrival time point 2 。
Preferably, in the first step, the calculation formula of the iterative cumulative sum of squares statistic is set as
Wherein T = T 2 -t 1 +1。
Preferably, in the third step, the calculation formula for generating the objective function of the antagonistic neural network model is
Preferably, in the fourth step, the forward data set, the middle data set and the near data set respectively account for 57.6%, 19.9% and 22.5% of the original data.
Compared with the prior art, the invention has the beneficial effects that: according to the training data generation method for EEW real-time p-wave collection, p-waves can be picked up in time only by one experience parameter of window time, the wrong setting risk is small, the problems of wrong picking and missed picking of the waves picked in time are avoided, and the picking is more reliable; the initial motion arrival time of the p-wave signal is picked up twice by using a long-time and short-time average value ratio method and an iterative accumulation square sum algorithm, so that the operation speed is high, the estimation error is small, the picking efficiency is high, and the result is more accurate; through the screening of the Pearson correlation coefficient algorithm and the interception of a specific time length, and the expansion of a generated countermeasure neural network, the correlation and the availability of data are improved, the data volume is sufficient, and the generated training data has higher quality.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, an embodiment of the present invention: a training data generation method for EEW real-time p-wave acquisition comprises the following steps: step one, screening a sample; estimating arrival time; step three, resisting expansion; step four, training and constructing;
in the first step, detecting a p-wave observation signal generated by an earthquake in real time through an earthquake monitoring station, drawing an observation acceleration diagram of the p-wave observation signal on a data window by taking time as a horizontal axis and acceleration as a vertical axis, then taking the acceleration as the data characteristic of the p-wave observation signal, applying a Pearson correlation coefficient algorithm, wherein the value range of the Pearson correlation coefficient is [ -1,1], the closer the absolute value of the correlation coefficient is to 1, the more relevant the two attributes are, the positive and negative values of the value represent the positive correlation or the negative correlation, the positive value represents the positive correlation, the negative value represents the negative correlation, and a large amount of historical data with strong characteristic correlation, namely p-wave historical signals, are screened from a database of the earthquake monitoring station and are used as a sample set of a real-time p-wave acquisition model of an EEW system;
in the second step, all samples in the sample set are processed by using a long-short time average ratio method, a rough arrival time point of the p-wave historical signal in each sample is picked up, an iterative cumulative sum-of-squares algorithm is applied, and a calculation formula of iterative cumulative sum-of-squares statistics is set as
Wherein T = T 2 -t 1 +1, intercepting the signal in specific time length area before and after the rough arrival time point, calculating the iterative cumulative square sum statistic and finding out the time point corresponding to the maximum value of the statistic, the time point being the accurate arrival time point of the p-wave historical signal, intercepting the signal in specific time length area before and after the accurate arrival time point as the original data of the real-time p-wave acquisition model of the EEW system, the specific time length area selecting t 5.12s before the rough arrival time point 1 T to 3s after the coarse arrival time point 2 ；
In the third step, the original data is used as a discrimination set to be input into a discriminator for generating the antagonistic neural network model, the randomly generated noise data is used as a generation set to be input into a generator for generating the antagonistic neural network model, the maximum and minimum game antagonistic training is carried out, and the calculation formula of the objective function for generating the antagonistic neural network model is
A large amount of expansion data which is very close to the discrimination set is generated by the confrontation;
in the fourth step, according to a recording time axis, original data are respectively split into a long-term data set, a medium-term data set and a short-term data set, the long-term data set, the medium-term data set and the short-term data set respectively account for 57.6%, 19.9% and 22.5% of the original data, then the extended data are equally split into three data sets, and the three data sets are respectively added into the long-term data set, the medium-term data set and the short-term data set to obtain a training data set, a verification data set and a test data set, wherein the training data set is training data of a real-time p-wave acquisition model of the EEW system.
Based on the above, the method has the advantages that the p-wave arrival selection can be performed only by one empirical parameter of window time, the wrong setting risk is small, the problems of wrong pick-up and missed pick-up of the pick-up in the arrival of the wave are solved, the pick-up is more reliable, the initial arrival time of the p-wave signal is picked up twice by using a long-time average value ratio method and an iterative cumulative sum of squares algorithm, the operation speed is high, the estimation error is small, the pick-up efficiency is high, the result is more accurate, in addition, the correlation and the usability of data are improved by screening and intercepting in a specific time length through a Pearson correlation coefficient algorithm and expanding a generated antagonistic neural network, the data quantity is sufficient, and the generated training data quality is higher.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (6)
1. A training data generation method for EEW real-time p-wave acquisition comprises the following steps: step one, screening a sample; step two, estimating arrival time; step three, resisting expansion; step four, training and constructing; the method is characterized in that:
in the first step, a p-wave observation signal generated by an earthquake is detected in real time through an earthquake monitoring station, an observation acceleration diagram of the p-wave observation signal is drawn on a data window by taking time as a horizontal axis and acceleration as a vertical axis, and then a large amount of historical data with strong characteristic correlation, namely the p-wave historical signal, is screened out from a database of the earthquake monitoring station by using a Pearson correlation coefficient algorithm and is used as a sample set of a real-time p-wave acquisition model of an EEW system by taking acceleration as data characteristics of the p-wave observation signal;
in the second step, all samples in the sample set are processed by using a long-short time average ratio method, a rough arrival time point of a p-wave historical signal in each sample is picked up, an iterative accumulation sum-of-squares algorithm is applied, signals in a specific time length region before and after the rough arrival time point are intercepted, iterative accumulation sum-of-squares statistics are calculated, a time point corresponding to the maximum value of the statistics is found out, the time point is an accurate arrival time point of the p-wave historical signal, and signals in the specific time length region before and after the accurate arrival time point are intercepted and serve as original data of a real-time p-wave acquisition model of the EEW system;
in the third step, the original data is taken as a discrimination set and input into a discriminator for generating a confrontation neural network model, then noise data generated randomly is taken as a generation set and input into a generator for generating the confrontation neural network model, and the extremely-small game confrontation training is carried out to generate a large amount of expansion data which is very close to the discrimination set;
in the fourth step, according to a recording time axis, original data is respectively split into a long-term data set, a medium-term data set and a short-term data set, then the expanded data is split into three data sets which are equally split, and the three data sets are respectively added into the long-term data set, the medium-term data set and the short-term data set to obtain a training data set, a verification data set and a test data set, wherein the training data set is training data of a real-time p-wave acquisition model of the EEW system.
2. A method of generating training data for EEW real-time p-wave acquisition according to claim 1, wherein: in the first step, the value range of the Pearson correlation coefficient is [ -1,1], the closer the absolute value of the correlation coefficient is to 1, the more relevant the two attributes are, the positive and negative values of the value represent the positive or negative correlation of the two attributes, the positive value represents the positive correlation, and the negative value represents the negative correlation.
3. A method of generating training data for EEW real-time p-wave acquisition as claimed in claim 1, wherein: in the second step, the specific time length region is selected 5 before the coarse arrival time point.T of 12s 1 T to 3s after the coarse arrival time point 2 。
4. A method of generating training data for EEW real-time p-wave acquisition as claimed in claim 1, wherein: in the first step, the calculation formula of the iterative cumulative sum of squares statistic is set as
Wherein T = T 2 -t 1 +1。
5. A method of generating training data for EEW real-time p-wave acquisition according to claim 1, wherein: in the third step, the calculation formula of the objective function for generating the antagonistic neural network model is
6. A method of generating training data for EEW real-time p-wave acquisition as claimed in claim 1, wherein: in the fourth step, the forward data set, the intermediate data set and the near data set respectively account for 57.6%, 19.9% and 22.5% of the original data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210938153.5A CN115438716A (en) | 2022-08-05 | 2022-08-05 | Training data generation method for EEW real-time p-wave acquisition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210938153.5A CN115438716A (en) | 2022-08-05 | 2022-08-05 | Training data generation method for EEW real-time p-wave acquisition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115438716A true CN115438716A (en) | 2022-12-06 |
Family
ID=84242634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210938153.5A Pending CN115438716A (en) | 2022-08-05 | 2022-08-05 | Training data generation method for EEW real-time p-wave acquisition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115438716A (en) |
-
2022
- 2022-08-05 CN CN202210938153.5A patent/CN115438716A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3468162B1 (en) | Method and device for tracking echo delay | |
| CN107479094B (en) | A method of realizing earthquake pre-warning | |
| CN111060965B (en) | Seismic facies picking and event detection method based on convolutional neural network | |
| CN106896407B (en) | A kind of microseism signal first break pickup method based on approximate negentropy | |
| CN104538041A (en) | Method and system for detecting abnormal sounds | |
| CN106052852B (en) | A kind of detection method and device of pulse acoustical signal | |
| CN111600876B (en) | Slow denial of service attack detection method based on MFOPA algorithm | |
| KR101468624B1 (en) | Terminal, system and method for measuring network state using the same | |
| CN109283576B (en) | Method for automatically picking up seismic phase of P wave by taking amplitude as characteristic function | |
| CN110018515A (en) | A method of with amplitude variations length hourly value than realizing P wave automatic Picking in earthquake pre-warning | |
| CN111126434B (en) | Random forest-based automatic pickup method and system for arrival time of microseismic first arrival | |
| CN110475246B (en) | Malicious anchor node detection method based on isolated forest and sequential probability ratio detection | |
| CN116828354B (en) | Radio quality optimization method and system for collar clamp wireless microphone | |
| US7916814B2 (en) | Method and apparatus for real-time pulse parameter estimator | |
| CN106382981A (en) | Single station infrasonic wave signal recognition and extraction method | |
| CN110398775B (en) | Tunnel water burst disaster micro-seismic event signal fluctuation first arrival pickup method and system | |
| CN105975995B (en) | More vibration signal fusion methods based on fuzzy preference relation | |
| CN110995713A (en) | Botnet detection system and method based on convolutional neural network | |
| CN115438716A (en) | Training data generation method for EEW real-time p-wave acquisition | |
| CN111505581B (en) | Passive target detection method based on distributed sensor nodes | |
| CN109655885A (en) | A kind of automatic first break pickup method and system | |
| CN117439827B (en) | Network flow big data analysis method | |
| CN115952410B (en) | Landslide hazard detection system based on deep learning | |
| CN115201826A (en) | Radar target intelligent track fusion method | |
| Jiang et al. | Non-line of sight error mitigation in UWB ranging systems using information fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |