CN-122019962-A - Method for constructing seismic interference signal rejection model, method and system for rejecting seismic interference signal

CN122019962ACN 122019962 ACN122019962 ACN 122019962ACN-122019962-A

Abstract

The invention discloses a method for constructing a seismic interference signal rejection model, a rejection method and a system thereof, relating to the field of geophysical data processing, comprising the steps of acquiring historical waveform data of a triggering event from a historical seismic triggering database; generating a training sample set based on the labeling result of the historical waveform data, inputting the training sample set into a long-short-term memory network, extracting the time sequence characteristics of the historical waveform data by using the long-short-term memory network to obtain the time sequence characteristics of the seismic signals, calculating the characteristic weight aggregation algorithm based on the time sequence characteristics of the seismic signals, calculating the prediction category of the historical waveform data by a classification decision algorithm based on the characteristic weight aggregation characteristics of the seismic signals, and finally performing iterative training on the long-short-term memory network based on the historical waveform data, the labeling category and the prediction category to obtain a trained seismic interference signal rejection model. The method and the device can effectively remove the interference signals and improve the processing reliability of the seismic data.

Inventors

LIN SEN
ZHU ZHUBING
LU ZHICHENG
GAO PO
SUN YUHAN
MENG XIANZHENG
XUE YAODONG

Assignees

国网电力工程研究院有限公司
国家电网有限公司
国网四川省电力公司电力科学研究院

Dates

Publication Date: 20260512
Application Date: 20251209

Claims (20)

1. The method for constructing the seismic interference signal rejection model is characterized by comprising the following steps of; Generating a training sample set based on the labeling result of the historical waveform data, wherein the training sample set comprises the historical waveform data and the corresponding labeling category thereof; the training sample set is input into a long-short-period memory network, the long-short-period memory network is utilized to extract time sequence characteristics of the historical waveform data to obtain time sequence characteristics of the seismic signals; and performing iterative training on the long-short-term memory network based on the historical waveform data, the labeling category and the prediction category to obtain a trained seismic interference signal rejection model.
2. The method of claim 1, wherein retrieving historical waveform data for the triggering event from a historical seismic trigger database comprises: Acquiring a plurality of triggered seismic event record data from a historical seismic trigger database of a historical seismic station network; The method comprises the steps of taking triggering time as a reference for each triggered seismic event record data, and adopting a time window with fixed length to intercept historical waveform data of a triggering event; wherein the historical waveform data includes a vertical component, a north-south component, and an east-west component.
3. The method of claim 1, wherein generating a training sample set based on labeling results of the historical waveform data comprises: performing examination and labeling on the historical waveform data, and labeling a corresponding seismic event category label on each time window data according to examination and labeling results to generate a training sample set; when the seismic event type label value is 1, the occurrence of the seismic event is indicated; And when the seismic event category label value is 0, indicating that a disturbance event occurs.
4. The method of claim 1, wherein prior to inputting the training sample set into a long and short term memory network, further comprising: Performing mean value removal and linear trend removal on all components in each historical waveform data in the training sample set, and applying a band-pass filter to perform noise removal processing to obtain each historical waveform data after the noise removal processing; And carrying out normalization processing on all components in each historical waveform data after the denoising processing.
5. The method of claim 1, wherein the performing time series feature extraction on the historical waveform data using the long-short term memory network to obtain the time series feature of the seismic signal comprises: and extracting the time sequence characteristics of the seismic signals of the historical waveform data in the time dimension by adopting a time sequence characteristic extraction module of the long-short-period memory network model, wherein the time sequence characteristics of the seismic signals are used for identifying the seismic signals.
6. The method of claim 1, wherein the computing the seismic signal aggregate signature using a signature weighted aggregate algorithm based on the seismic signal timing signature comprises: Based on the time sequence characteristics of the seismic signals, attention weight at each moment in the time sequence characteristics of the seismic signals is calculated by adopting an attention mechanism algorithm; and calculating the seismic signal aggregation characteristics by adopting a characteristic weighting aggregation module of the long-short-period memory network model based on the attention weight and the seismic signal time sequence characteristics at each moment.
7. The method of claim 6, wherein the feature weighted aggregation module has a functional expression of: Wherein, the Attention weight at time t; the hidden state is the t moment, and Z is the seismic signal aggregation characteristic.
8. The method of claim 1, wherein the calculating a predicted category of the historical waveform data by a classification decision algorithm based on the seismic signal aggregate characteristics comprises: Performing nonlinear activation processing on the seismic signal aggregation characteristics through a full-connection layer of the long-short-term memory network model to obtain seismic characteristic space data; And determining the prediction category of the historical waveform data by adopting a classification decision module of the long-short-term memory network model based on the seismic feature space data.
9. A seismic disturbance signal rejection model building system, comprising: the data acquisition module is used for acquiring historical waveform data of the triggering event from the historical earthquake triggering database; The marking module is used for generating a training sample set based on the marking result of the historical waveform data, wherein the training sample set comprises the historical waveform data and the corresponding marking category thereof; the time sequence feature extraction module is used for inputting the training sample set into a long-short-period memory network, and extracting time sequence features of the historical waveform data by utilizing the long-short-period memory network to obtain time sequence features of the seismic signals; the characteristic aggregation module is used for calculating the seismic signal aggregation characteristics by adopting a characteristic weighting aggregation algorithm based on the time sequence characteristics of the seismic signals; The classification decision module is used for obtaining the prediction category of the calculated historical waveform data through a classification decision algorithm based on the seismic signal aggregation characteristics; and the training module is used for carrying out iterative training on the long-period memory network based on the historical waveform data, the labeling category and the prediction category to obtain a trained seismic interference signal rejection model.
10. The system of claim 9, wherein the data acquisition module is specifically configured to acquire a plurality of triggered seismic event record data from a historical seismic trigger database of a historical seismic platform network; The method comprises the steps of taking triggering time as a reference for each triggered seismic event record data, and adopting a time window with fixed length to intercept historical waveform data of a triggering event; wherein the historical waveform data includes a vertical component, a north-south component, and an east-west component.
11. The system of claim 9, wherein the labeling module is specifically configured to perform examination labeling on the historical waveform data, label a corresponding seismic event category label for each time window data according to an examination result, and generate a training sample set; when the seismic event type label value is 1, the occurrence of the seismic event is indicated; And when the seismic event category label value is 0, indicating that a disturbance event occurs.
12. The system of claim 9, wherein the system further comprises: The data preprocessing module is used for carrying out mean value removal and linear trend removal processing on all components in each historical waveform data in the training sample set, and applying a band-pass filter to carry out noise removal processing; And carrying out normalization processing on all components in each historical waveform data after the denoising processing.
13. The system of claim 9, wherein the timing feature extraction module is specifically configured to extract a timing feature of the seismic signal in a time dimension thereof using a long-short term memory network model timing feature extraction module, the timing feature of the seismic signal being configured to identify the seismic signal.
14. The system of claim 9, wherein the feature aggregation module is specifically configured to calculate, based on the seismic signal timing features, a attention weight at each moment in the seismic signal timing features using an attention mechanism algorithm; and calculating the seismic signal aggregation characteristic by adopting a characteristic weighting aggregation module based on the attention weight and the seismic signal time sequence characteristic.
15. The system of claim 14, wherein the feature weighted aggregation module has a functional expression of: Wherein, the Attention weight at time t; the hidden state is the t moment, and Z is the seismic signal aggregation characteristic.
16. The system of claim 9, wherein the classification decision module is specifically configured to perform nonlinear activation processing on the seismic signal aggregate features through a fully connected layer to obtain seismic feature spatial data; Based on the seismic feature space data, a classification decision module is adopted to calculate the prediction category of the historical waveform data according to the classification type of the seismic signals.
17. An interference signal rejection method based on a seismic interference signal rejection model is characterized by comprising the following steps of; When a triggering event report is sent out by a seismic station, acquiring current waveform data from the triggering event report; Performing interference signal rejection on the seismic waveform data by utilizing a pre-constructed seismic interference signal rejection model to obtain event categories corresponding to the current waveform data; Judging whether the triggering event report is a real earthquake event or not based on the event category, and triggering earthquake early warning if the triggering event report is a real earthquake event; the seismic interference signal rejection model is constructed by a seismic interference signal rejection model construction method according to any one of claims 1 to 8.
18. The method of claim 17, wherein the determining whether the trigger event report is a real seismic event based on the event category comprises: When the triggering probability of the event category is larger than a preset threshold value, judging that the triggering event is reported as a real earthquake event; And when the triggering probability of the event category is not greater than a preset threshold value, judging that the triggering event is reported as an interference event.
19. An interfering signal rejection system based on a seismic interfering signal rejection model, comprising: The data acquisition module is used for acquiring current waveform data from a triggering event report when the triggering event report is sent out by the earthquake station; the interference signal rejection module is used for rejecting the interference signal of the seismic waveform data by utilizing a pre-constructed seismic interference signal rejection model to obtain event categories corresponding to the current waveform data; the earthquake early warning module is used for judging whether the triggering event report is a real earthquake event or not based on the event category, and triggering earthquake early warning if the triggering event report is a real earthquake event; Wherein the seismic disturbance signal rejection model is constructed by a seismic disturbance signal rejection model construction method according to any one of claims 1 to 8.
20. The system of claim 19, wherein the earthquake early warning module is specifically configured to, when the triggering probability of the event category is greater than a preset threshold, determine that the triggering event is reported as a real earthquake event, and trigger an earthquake early warning; and when the triggering probability of the event category is not greater than a preset threshold value, judging that the triggering event is reported as an interference event, and not triggering earthquake early warning.

Description

Method for constructing seismic interference signal rejection model, method and system for rejecting seismic interference signal Technical Field The invention relates to the technical field of geophysical data processing, in particular to a method for constructing a seismic interference signal rejection model, a rejection method and a system. Background At present, in an earthquake monitoring and automatic cataloging system, the identification and classification of event signals are key links for ensuring data quality and early warning accuracy. Conventional seismic event discrimination methods typically rely on artificial feature extraction in combination with technical routes of conventional machine learning algorithms. Specifically, researchers manually extract a series of characteristic parameters from the seismic waveform, such as peak frequency, frequency spectrum width, P/S wave energy ratio, bulk wave magnitude to surface wave magnitude ratio method, envelope morphology, polarization characteristics and the like, and then judge the seismic event and the non-seismic interference signals by using classification models such as a Support Vector Machine (SVM), a random forest or a shallow neural network and the like. The method has the following objective defects that (1) the dependence of the algorithm on artificial characteristic engineering is strong, different types of interference signals (such as traffic vibration, blasting, instrument pulse and the like) have obvious differences in waveform characteristics, and the fixed manual characteristics are difficult to cover all interference modes, so that the generalization capability of a model is weak. (2) The feature extraction process requires a large amount of knowledge and manual participation in the seismology field, and has higher design and maintenance cost, thereby being unfavorable for popularization and application in a large-scale automatic system. (3) The traditional method classifies based on statistical characteristics, and is difficult to fully reflect dynamic evolution characteristics (such as successive occurrence rules of P waves, S waves and wake waves) of waveform signals in time dimension, so that the recognition accuracy of a model on complex signals is limited. In summary, the conventional earthquake early warning event discriminating method generally relies on artificial feature extraction and machine learning algorithm, so that it is difficult to effectively capture the time sequence dynamic features of the earthquake waveform, and the problem of the recognition accuracy of the model to the complex signal is limited. Disclosure of Invention In order to solve the problems that the time sequence dynamic characteristics of the earthquake waveform are difficult to effectively capture and the recognition accuracy of the model to the complex signals is limited in the earthquake early warning event in the prior art. In a first aspect, the present invention provides a method for constructing a seismic interference signal rejection model, including; Generating a training sample set based on the labeling result of the historical waveform data, wherein the training sample set comprises the historical waveform data and the corresponding labeling category thereof; the training sample set is input into a long-short-period memory network, the long-short-period memory network is utilized to extract time sequence characteristics of the historical waveform data to obtain time sequence characteristics of the seismic signals; and performing iterative training on the long-short-term memory network based on the historical waveform data, the labeling category and the prediction category to obtain a trained seismic interference signal rejection model. Preferably, acquiring historical waveform data of the triggering event from a historical seismic triggering database includes: Acquiring a plurality of triggered seismic event record data from a historical seismic trigger database of a historical seismic station network; The method comprises the steps of taking triggering time as a reference for each triggered seismic event record data, and adopting a time window with fixed length to intercept historical waveform data of a triggering event; wherein the historical waveform data includes a vertical component, a north-south component, and an east-west component. Preferably, the generating a training sample set based on the labeling result of the historical waveform data includes: performing examination and labeling on the historical waveform data, and labeling a corresponding seismic event category label on each time window data according to examination and labeling results to generate a training sample set; when the seismic event type label value is 1, the occurrence of the seismic event is indicated; And when the seismic event category label value is 0, indicating that a disturbance event occurs. Preferably, before the training sample set is input into the