CN-121995486-A - Near-field P wave plate segment driven earthquake disaster analysis method and system

CN121995486ACN 121995486 ACN121995486 ACN 121995486ACN-121995486-A

Abstract

The invention discloses a near-field P wave plate segment driven earthquake disaster analysis method and a near-field P wave plate segment driven earthquake disaster analysis system, wherein the method extracts multidimensional fusion characteristics of preprocessed earthquake waveform data through connection of a multi-window convolution parallel architecture and residual errors, and the multi-window convolution parallel architecture adopts convolution kernels with different sizes to respectively capture characteristics of P waves in a designated period; the method comprises the steps of inputting multidimensional fusion characteristics into a parallel integrated encoder, obtaining internal time sequence association of waveforms through a self-attention mechanism, adaptively distributing different weights to characteristics of a designated period, and outputting the waveform characteristics through global average pooling, inputting the global characteristics into a multi-path decoder branch, and outputting earthquake disaster analysis parameters comprising initial motion polarity, earthquake type, earthquake magnitude, waveform quality and risk level probability. The method solves the problems of insufficient real-time performance, waveform quality evaluation distortion, poor disaster-causing classification accuracy and poor multi-task cooperativity in the prior art.

Inventors

Jia Leizhao
CHEN SHI
WU YAFENG
XING KANG
LI YONGBO
WANG FEIFEI

Assignees

河南省地震局

Dates

Publication Date: 20260508
Application Date: 20260126

Claims (10)

1. The earthquake disaster analysis method driven by the near-field P wave plate segment is characterized by comprising the following steps of: Receiving near-field P-wave seismic waveform data recorded by multiple channels of multiple stations, and performing trending, band-pass filtering and normalization pretreatment on the waveform data; Extracting multiscale fusion characteristics of the preprocessed waveform data through connection of a multiscale convolution parallel architecture and residual errors, wherein the multiscale convolution parallel architecture adopts convolution kernels with different sizes to respectively capture characteristics of a P wave in a specified period; Inputting the multi-scale fusion features into a parallel integrated encoder, acquiring waveform internal time sequence association through a self-attention mechanism, adaptively distributing different weights to the features of a designated time period, and outputting waveform global features through global average pooling; And inputting the global features into a multi-path decoder branch, and outputting earthquake disaster analysis results comprising initial motion polarity, earthquake type, earthquake magnitude, waveform quality and risk level probability.
2. The near-field P-wave segment driven seismic disaster analysis method of claim 1 wherein the duration of the near-field P-wave seismic waveform data is 10 seconds, the band of the band pass filter is 1-20Hz, the normalization preprocessing removes baseline drift trend of the waveform by linear fitting, and then linear scaling normalization is adopted.
3. The near-field P-wave plate segment driven earthquake disaster analysis method of claim 1, wherein in the multi-window convolution parallel architecture, convolution kernel sizes comprise three types of 3,5 and 7, a convolution kernel with the size of 3 captures abrupt features of a P-wave initial motion segment, a convolution kernel with the size of 5 captures main frequency features of a vibration stationary segment, and a convolution kernel with the size of 7 captures energy change features of an attenuation segment; And the multi-scale fusion characteristics are input into a parallel integrated encoder after being subjected to 1X 1 convolution dimension reduction treatment, and the dimension of the characteristics after dimension reduction is 320 dimensions.
4. The near-field P-wave segment driven earthquake disaster recovery analysis method of claim 1, wherein the number of layers of the parallel integrated encoder is 4, each layer of the parallel integrated encoder comprises a multi-head self-attention mechanism and a feed-forward network, and the number of heads of the multi-head self-attention mechanism is 8; The dimension of the global feature is 320 dimensions, and the global feature comprises the initial motion section amplitude direction of the seismic wave, the sampling point level amplitude mutation rule, the waveform signal to noise ratio, the baseline stability, the amplitude change trend, the peak amplitude, the overall attenuation rate and the energy distribution rule.
5. The near field P-wave segment driven seismic disaster analysis method of claim 1 wherein the multi-path decoder branches comprise a primary polarity decoder, a seismic type decoder, a magnitude decoder, a waveform quality decoder, and a risk level decoder; the initial motion polarity decoder adopts a full-connection structure, and outputs positive polarity probability distribution or negative polarity probability distribution through a Softmax activation function; the earthquake type decoder adopts a full-connection structure, and outputs natural earthquake, artificial blasting or collapse type probability through a Softmax activation function; the waveform quality decoder adopts a full-connection structure and outputs probability distribution of high, medium and low quality grades through a Softmax activation function; the risk level decoder adopts a fully-connected structure, and outputs low, medium, high and extremely high risk probability distribution through a Softmax activation function.
6. The near-field P-wave plate segment driven earthquake disaster analysis method of claim 5, further comprising the steps of adaptively weighting and aggregating a plurality of station data through waveform quality, and optimizing event-level global index output by combining a two-way calibration mechanism of station-level and event-level results; the quality assessment of the waveform is realized through triple decoupling constraint, the method comprises the steps of eliminating energy difference through input layer standardization processing, and avoiding quality assessment to depend on energy characteristics through energy parameter influence isolation by a feature layer and adding a relevance penalty term by a loss layer through an attention mask; The waveform quality calculates the total mass score based on the weighting of four indexes of signal-to-noise ratio, baseline drift degree, amplitude mutation rate and P-wave integrity And determining, namely, the specific calculation mode is as follows: , in the formula, For the normalized signal-to-noise ratio, For the normalized baseline drift level, For the normalized amplitude mutation rate, Is normalized P-wave integrity.
7. The near-field P-wave segment driven earthquake disaster analysis method of claim 6, wherein the four indexes are normalized by: , , , , the signal-to-noise ratio SNR is the root mean square ratio of the amplitude of a signal section to the noise section in the 10-second P-wave waveform, the signal section is 8 seconds after the first arrival of the P-wave in the 10-second waveform, and the noise section is 2 seconds before the first arrival of the P-wave; The baseline drift BD is the product of the absolute value of the slope of the waveform linear fitting straight line after pretreatment and the number of sampling points; The amplitude mutation rate AMR is the proportion of the times that the amplitude difference value of adjacent sampling points in the waveform exceeds the integral amplitude standard deviation by 3 times to the total sampling points; the P wave integrity PI identifies the P wave arrival time through an energy detection method, if the P wave arrival time appears in the first 3 seconds of the 10-second waveform, PI=1, otherwise, PI=0.
8. The near field P-wave segment driven seismic disaster analysis method of claim 6 wherein the risk level probabilities are generated by risk index calculation, the risk index Based on the magnitude estimation value, the epicenter distance and the field type, the concrete calculation formula is as follows: , Wherein M is a magnitude measuring and calculating value; is the linear distance between the station and the seismic center; the method is characterized in that the method is used for identifying the field type and is divided into hard rock and soft soil according to the shear wave speed; the risk score for disaster is as follows: When 0 is When <1, the risk level is low risk, and the tag value is 0; When 1 When the risk grade is the medium risk, the label value is 1; When 3 is When the risk grade is high risk, the label value is 2; when 5 When the risk level is extremely high, the label value is 3; the bi-directional calibration mechanism includes: The risk level of all stations is adjusted down to low risk when the event type is artificial blasting, the risk level of the adjacent stations is calibrated to high risk when the event type is collapse, the vibration level re-estimation is triggered when more than or equal to 80% of stations output high risk but the event level vibration level is less than 2.5, and the event level result is marked to be rechecked when the waveform quality of more than or equal to 90% of stations is judged to be low and reliable.
9. The near-field P-wave plate segment driven earthquake disaster analysis method of claim 1, further comprising a double-layer task cooperative training mechanism, wherein the loss weight of the station level at the initial stage of training is 0.7, the loss weight of the station level at the event stage is 0.3, the loss weight of the station level at the later stage of training is 0.3, and the loss weight of the station level at the event stage is 0.7; The classification task adopts cross entropy loss, the regression task adopts Huber loss, the semi-supervision consistency loss is additionally added, the adaptation part labels for missing, the network weight is reversely updated through a AdamW optimizer in the training process, and the total loss is obtained by weighting and summing the weight ratio of the polarity loss, the earthquake type loss, the magnitude loss, the waveform quality loss and the risk level loss to 0.15:0.2:0.25:0.2:0.2.
10. A near-field P-wave segment driven earthquake disaster analysis system employing the near-field P-wave segment driven earthquake disaster analysis method as set forth in any one of claims 1 to 9, characterized by comprising: The data preprocessing module is used for receiving near-field P-wave seismic waveform data recorded by multiple channels of multiple stations and carrying out trending, band-pass filtering and normalization preprocessing on the waveform data; The characteristic extraction module is used for extracting multiscale fusion characteristics of the preprocessed waveform data through connection of a multiscale convolution parallel architecture and residual errors, and the multiscale convolution parallel architecture adopts convolution kernels with different sizes to respectively capture the characteristics of the P wave in a specified period; The shared coding module is used for inputting the multi-scale fusion characteristics into the parallel integrated coder, acquiring the internal time sequence association of the waveform through a self-attention mechanism, adaptively distributing different weights to the characteristics of a specified period, and outputting the global characteristics of the waveform through global average pooling; The multi-task decoding module is used for inputting the global features into a multi-path decoder branch and outputting earthquake disaster analysis results comprising initial motion polarity, earthquake type, earthquake magnitude, waveform quality and risk level probability; and the aggregation calibration module is used for adaptively weighting and aggregating a plurality of station data through waveform quality and optimizing the output of the event-level global index by combining a bidirectional calibration mechanism of the station-level and event-level results.

Description

Near-field P wave plate segment driven earthquake disaster analysis method and system Technical Field The invention relates to the technical field of seismic monitoring, in particular to a near-field P-wave plate segment driven seismic disaster analysis method and system. Background In the fields of earthquake monitoring, explosion monitoring, mine safety monitoring and the like, monitoring and early warning of natural earthquakes, collapse, industrial blasting and other non-natural earthquake events are the requirements for guaranteeing personnel and property safety and improving disaster prevention and reduction capability. The method has the advantages of quickly acquiring the seismic parameters, accurately identifying the event types and scientifically judging the disaster causing grade, directly relates to the emergency response efficiency and the effectiveness of disaster prevention decision, and has important application value in the technical fields of seismic monitoring, geological exploration, mineral production safety and the like. In the aspects of seismic parameter acquisition and event identification, the current mainstream seismic monitoring technical scheme has a remarkable short plate. The prior art generally relies on complete seismic wave sequences, and needs to wait for the completion of recording and acquisition of P waves, S waves and other full time periods to start research and judgment, so that monitoring delay reaches a minute level, and the real-time requirement of short-term early warning is difficult to meet. The partial short P wave analysis technology has the problems of simple structure, single parameter output, lack of waveform quality pre-evaluation and the like, is easy to cause parameter estimation deviation due to low-quality waveforms, and has higher error rate in judging the type of the medium microseism. Waveform quality assessment is a key for guaranteeing parameter accuracy, but the existing method is mainly used for constructing a judging system based on absolute energy indexes, and is strongly coupled with energy parameters such as magnitude, middle distance and the like, so that quality assessment distortion is easy to cause. Meanwhile, the fixed threshold value has poor generalization capability, is difficult to adapt to the difference between different station instruments and regional geological media, cannot provide stable quality screening basis, and the parameter estimation error can be further amplified by mixing low signal-to-noise ratio signals. In the aspects of disaster-causing classification and disaster prevention decision, the existing scheme has obvious limitations. Most of the schemes only take the magnitude as a grading basis, neglect the regulation and control effects of the magnitude, geological conditions and the like on disasters, and few schemes introducing the magnitude also have the problems of rough grading and no accurate combination judgment logic, and are disjointed with the actual disaster prevention requirements. In addition, the prior art does not correlate waveform quality with disaster causing classification, is easy to trigger invalid early warning, outputs a result which is mostly abstract, can be subjected to butt joint disaster prevention decision only by manual secondary conversion, and has insufficient practicability. In the design of a model architecture, the mainstream single-task independent architecture has the problems of high deployment cost, low reasoning efficiency, poor consistency of multi-task results and the like. A few of the multi-task models do not design special structures aiming at the time sequence characteristics of the earthquake waves, the P-wave key time sequence associated characteristics are difficult to capture, a complete technical chain from basic parameters to quality evaluation to risk classification is not formed, and part of the related technologies also have the limitations of dependence on long-period waveforms, single parameter output and the like, so that the actual combat requirements of the non-natural earthquake early warning cannot be met. Disclosure of Invention Therefore, the invention provides a near-field P wave plate segment driven earthquake disaster analysis method and a near-field P wave plate segment driven earthquake disaster analysis system, which solve the problems of insufficient real-time performance, waveform quality evaluation distortion, disaster classification accuracy deficiency and poor multi-task cooperativity in the prior art. In order to achieve the purpose, the invention provides the technical scheme that the earthquake disaster analysis method driven by the near-field P wave plate segment comprises the following steps of: Receiving near-field P-wave seismic waveform data recorded by multiple channels of multiple stations, and performing trending, band-pass filtering and normalization pretreatment on the waveform data; extracting multiscale fusion characteristics