CN-122017968-A - Lightweight DAS (DAS) seismic phase pickup method and system based on sparse connection

CN122017968ACN 122017968 ACN122017968 ACN 122017968ACN-122017968-A

Abstract

The application discloses a lightweight DAS (data acquisition system) seismic facies pickup method and system based on sparse connection, which relate to the technical field of seismic exploration and comprise the steps of firstly segmenting DAS data of a target area, dividing local time windows, dynamically adjusting morphological structural elements based on each local time window to generate an enhanced signal characteristic sequence, inputting a parallel fusion network formed by convolution and attention mechanisms and a sparsely connected U-shaped network into the sequence, fusing multichannel characteristic graphs output by the two types of networks, correspondingly matching, weighting and superposing local and global characteristics, forming P wave and S wave arrival distribution curves reflecting time and space distribution, effectively reducing network redundancy through a sparse connection mechanism, improving calculation efficiency, simultaneously combining local convolution response and global attention information, realizing accurate capture of concentrated positions and propagation trends of P wave and S wave energy, and simultaneously guaranteeing complete characterization of a key vibration mode by a model under the condition of low calculation amount.

Inventors

DU SHUAIQUN
XIN JUNSHENG
ZHANG HENG

Assignees

中国雅江集团有限公司
中国科学院青藏高原研究所

Dates

Publication Date: 20260512
Application Date: 20251229

Claims (10)

1. The lightweight DAS seismic phase pickup method based on sparse connection is characterized by comprising the following steps of: Performing segmentation analysis on target area DAS data to obtain the length of each segmentation window of the target area DAS data, segmenting the target area DAS data to form a target area DAS signal sequence, dividing the target area DAS signal sequence to obtain each local time window, and dynamically adjusting morphological structural elements based on each local time window; forming an enhanced signal characteristic sequence according to a target area DAS signal subjected to the action of morphological structural elements, and simultaneously inputting the enhanced signal characteristic sequence into a parallel fusion network formed by convolution and an attention mechanism and a sparse connection U-shaped network; Accurately characterizing and identifying the vibration mode of the target area according to a parallel fusion network formed by convolution and an attention mechanism; constructing and monitoring multi-channel feature graphs corresponding to all convolutions and downsampling layers in convolution branches of a sparse connection U-shaped network, simultaneously analyzing channel activation intensity of the multi-channel feature graphs corresponding to all convolutions and downsampling layers, and judging the operation of convolutions and downsampling; And fusing a parallel fusion network formed by convolution and an attention mechanism and a multi-channel feature map output by a sparse connection U-shaped network, judging an arrival candidate area of the P wave and the S wave, and forming an arrival distribution curve of the P wave and the S wave.
2. The sparse-connection-based lightweight DAS seismic facies pickup method of claim 1, wherein the segmenting the target area DAS data into target area DAS signal sequences comprises the following steps: Performing segmentation analysis on target area DAS data, performing preliminary segmentation on the target area DAS data to obtain first segmentation windows of the target area DAS data, and obtaining time sequence vibration response data, including acoustic energy intensity, signal instantaneous change rate and signal waveform variance, of optical fibers along the lines in the first segmentation windows of the target area DAS data; The method comprises the steps of fusing the sound wave energy intensity, the signal instantaneous change rate and the signal waveform variance to obtain first dividing window scale adjustment factors of target area DAS data, wherein the first dividing window scale adjustment factors of the target area DAS data are used for evaluating the activity degree of signals in the first dividing windows, obtaining length time of each dividing window of the target area DAS data according to the first dividing window scale adjustment factors of the target area DAS data, and adjusting each first dividing window of the target area DAS data according to the length time of each dividing window of the target area DAS data, so that the target area DAS data are segmented to form a target area DAS signal sequence.
3. The sparse connection-based lightweight DAS seismic facies pickup method of claim 1, wherein the dynamically adjusting morphological structural elements based on each local time window comprises the following steps: Dividing a target area DAS signal sequence formed by segmenting target area DAS data into a plurality of local fragments according to time to obtain local time windows, extracting time sequence vibration response data of optical fibers of the target area DAS data in the local time windows to obtain scale adjustment factors of second local time windows of the target area DAS data, wherein the scale adjustment factors of the second local time windows of the target area DAS data are used for evaluating vibration mode stability and energy distribution change characteristics of local signals under fine granularity time scales, and dynamically adjusting morphological structural elements according to the scale adjustment factors of the second local time windows of the target area DAS data.
4. The sparse-connection-based lightweight DAS seismic facies pickup method of claim 1, wherein the accurately characterizing and identifying the vibration modes of the target region comprises the following steps: The method comprises the steps of arranging target area DAS enhancement signals obtained after morphological structural element processing according to an original time sequence to form a continuous enhancement signal characteristic sequence, inputting the enhancement signal characteristic sequence into a convolution branch, setting a plurality of convolution kernel groups with different lengths in the convolution branch, enabling each convolution kernel to slide along a time axis on the sequence according to a specific time scale, taking signal fragments of a current position and a neighborhood thereof at each sliding position, multiplying and summing the signal fragments with convolution kernel weights element by element to obtain local characteristic response values of each convolution kernel at the current position, wherein the local characteristic response values of each convolution kernel at the current position are used for evaluating the matching degree of local characteristics of target area DAS signals and convolution kernel weight modes in the time point and the neighborhood thereof; Acquiring local energy and waveform change data of signals in a signal interval covered by each convolution kernel when each convolution kernel slides to a certain time point, wherein the local energy and waveform change data comprise an average value of the square of the amplitude of a DAS signal in a target area and the instantaneous change rate of the signals; If the average value of the square of the signal amplitude of the DAS in the target area and the instantaneous change rate of the signal are both beyond the square interval of the corresponding signal amplitude and the instantaneous change rate interval of the signal stored in the database, the local characteristic response value of the convolution kernel at the current position corresponding to the signal interval covered by the convolution kernel when the convolution kernel slides to a certain time point is corrected, otherwise, the local characteristic response value of the convolution kernel at the current position is not required to be corrected.
5. The sparse-connection-based lightweight DAS seismic facies pickup method of claim 1, wherein the channel activation intensity analysis is performed on the multi-channel feature map corresponding to each convolution and downsampling layer, and the specific process is as follows: inputting a continuous enhanced signal characteristic sequence into a convolution branch of a sparse connection U-shaped network, performing convolution and downsampling operation in a coding path of the U-shaped network, extracting space-time characteristic response of each scale layer by layer to generate a multi-characteristic response matrix, automatically constructing a corresponding multi-channel characteristic map at each downsampling layer, and marking the multi-channel characteristic map as a multi-channel characteristic map corresponding to each downsampling layer; Extracting the feature vectors of each channel in the feature images corresponding to the automatic construction of the downsampling layers, taking the amplitude values of the feature vectors of each channel in the feature images point by point in the time dimension, accumulating and summing the amplitude values, dividing the sum by the total number of the feature vectors in the channels, and obtaining the average response value of each channel corresponding to each downsampling layer; Comparing the average response value of each channel corresponding to each downsampling layer with the average response threshold value of the channel corresponding to the downsampling layer stored in the database, if the average response value of a channel corresponding to a downsampling layer is lower than the average response threshold value of the channel corresponding to the downsampling layer, marking the channel corresponding to the downsampling layer as a low response channel, and not reserving the characteristics of the channel in jump connection, otherwise, marking the channel corresponding to the downsampling layer as a low response channel.
6. The sparse-connection-based lightweight DAS seismic phase pickup method of claim 5, wherein the determining of performing convolution and downsampling operations comprises: respectively carrying out statistical analysis on the multi-channel feature map corresponding to each downsampling layer in the time dimension and the channel dimension to obtain global statistics of each downsampling layer, wherein the global statistics comprise channel mean value, channel variance, time energy variance and time frequency energy entropy; Comparing the comprehensive complexity index value of each downsampling layer with the comprehensive complexity index threshold value of the downsampling layers stored in the database, automatically stopping further convolution and downsampling operation if the comprehensive complexity index value of a certain downsampling layer is lower than the comprehensive complexity index threshold value of the downsampling layer, and continuously executing further convolution and downsampling operation if the comprehensive complexity index value of a certain downsampling layer is higher than or equal to the comprehensive complexity index threshold value of the downsampling layer.
7. The sparse-connection-based lightweight DAS seismic facies pickup method of claim 1, wherein the fusing of parallel fusion networks composed of convolution and attention mechanisms and multi-channel feature graphs output by sparse-connection U-shaped networks comprises the following specific processes: Arranging the multichannel characteristic diagrams output by the convolution and attention mechanism parallel fusion network according to the time dimension and the channel dimension to obtain a local vibration mode response sequence, and meanwhile, arranging the multichannel characteristic diagrams output by the sparse connection U-shaped network according to the same time and channel dimension to obtain a multi-scale global vibration mode sequence; and carrying out point-by-point corresponding matching on the local vibration mode response sequence and the multi-scale global vibration mode sequence, and carrying out linear superposition processing to obtain a fused multi-channel characteristic sequence.
8. The sparse-connection-based lightweight DAS seismic facies pickup method of claim 4, wherein the determining of arrival candidate areas of P-waves and S-waves comprises: scanning the fused multi-channel characteristic sequence point by point along a time axis to obtain the amplitude of each channel at each time point; Presetting monitoring time periods, extracting local characteristic response values of the convolution kernels at the current position in each monitoring time period, forming a local characteristic response sequence on a time axis by the local characteristic response values of the convolution kernels at the current position, and averaging the amplitude values of the local characteristic response sequences to obtain global amplitude values of the monitoring time periods; And comparing the global amplitude of each monitoring time period with P-wave and S-wave typical amplitude thresholds stored in a database in advance, judging the frequency range of the local characteristic response sequence, and then carrying out joint marking with the optical fiber channel to form candidate P-wave and S-wave arrival areas containing time intervals and channel information.
9. The sparse-connection-based lightweight DAS seismic phase pickup method of claim 8, wherein the forming of the P-wave and S-wave arrival profile comprises: The time periods marked as candidate P wave and S wave arrival intervals in each monitoring time period and the corresponding amplitude information thereof are arranged in time sequence, a channel amplitude matrix is constructed by combining the spatial distribution positions of all the optical fiber channels, normalized mapping is carried out on the time axis according to the amplitude intensity, and the areas with higher amplitude are highlighted to form the P wave and S wave arrival three-dimensional distribution curve.
10. A system applying the sparse-connection-based lightweight DAS seismic phase pickup method of any one of claims 1-9, comprising: The DAS signal segmentation and local time window generation module is used for carrying out segmentation analysis on target area DAS data to obtain the length of each segmentation window of the target area DAS data, segmenting the target area DAS data to form a target area DAS signal sequence, dividing the target area DAS signal sequence to obtain each local time window, and dynamically adjusting morphological structural elements based on each local time window; the enhanced signal characteristic sequence generation and dual-network input module is used for forming an enhanced signal characteristic sequence according to target area DAS signals subjected to the action of morphological structural elements and simultaneously inputting the enhanced signal characteristic sequence into a parallel fusion network formed by convolution and an attention mechanism and a sparse connection U-shaped network; The parallel fusion network local and global vibration mode characterization module is used for accurately characterizing and identifying the vibration mode of the target area according to a parallel fusion network formed by convolution and an attention mechanism; The sparse connection U-shaped network feature map construction and channel activation monitoring module is used for constructing and monitoring multi-channel feature maps corresponding to all convolution and downsampling layers in convolution branches of the sparse connection U-shaped network, simultaneously carrying out channel activation intensity analysis on the multi-channel feature maps corresponding to all convolution and downsampling layers, and judging the operation of executing the convolution and downsampling; And the multichannel characteristic fusion judging module is used for fusing a parallel fusion network formed by convolution and an attention mechanism and a multichannel characteristic diagram output by a sparse connection U-shaped network, judging the arrival candidate areas of the P wave and the S wave and forming an arrival distribution curve of the P wave and the S wave.

Description

Lightweight DAS (DAS) seismic phase pickup method and system based on sparse connection Technical Field The invention relates to the technical field of seismic exploration, in particular to a lightweight DAS (distributed system) seismic phase pickup method and system based on sparse connection. Background In modern seismic exploration and geological monitoring, distributed fiber optic acoustic sensing (DAS) technology has received attention because of its ability to achieve large-scale, high-density, continuous acquisition of vibratory signals along fiber optic lines. The DAS system can convert common optical fibers into a high-density sensing array, acquire sound waves or vibration information along the line through demodulation of phase or frequency changes, and provide abundant data for seismic wave arrival detection, geologic structure analysis and surface and underground activity monitoring. However, DAS signals are generally accompanied by high noise background, non-ideal response of optical fibers and signal attenuation problems, meanwhile, the data volume is extremely large, and the traditional seismophase pickup method based on a fully-connected deep neural network or a complex convolution network is high in calculation cost and storage requirement when processing large-scale DAS data, is easily interfered by redundancy features and local noise, and causes the reduction of the recognition precision of P wave arrival time and S wave arrival time. In this case, DAS data acquired along the optical fiber typically features local energy concentration, significant phase fluctuations, and non-uniform spatial distribution, with large differences in signal amplitude across different channels and time periods. The external interference, the geological condition change and the difference of the optical fiber laying state can also cause the non-uniformity of the signal distribution, so that the traditional signal analysis method is difficult to accurately identify the arrival time and the spatial distribution of the P wave and the S wave. Meanwhile, the existing method has large calculation amount and high storage consumption when processing large-scale DAS data, is sensitive to noise and local abnormal signals, and is difficult to realize real-time and efficient data processing while ensuring the precision. Therefore, in DAS optical cable monitoring, the influence caused by the local and global characteristics, the complex propagation effect and the optical cable laying mode of signals needs to be fully considered, and reasonable preprocessing and feature analysis are performed on DAS data so as to solve the problems of waveform complexity and interference, thereby providing a technical foundation for high-precision and high-reliability vibration phase identification and waveform analysis. The technical problems to be solved are that the method for picking up the seismic facies in time based on a space-time attention mechanism is provided, the method for picking up the seismic facies in time based on the space-time attention mechanism is improved, the technical scheme is adopted, the method comprises the steps of obtaining seismic signal data, marking P wave and S wave of the data, preprocessing the seismic signal data, preprocessing the data, including data enhancement, data set segmentation and spectrogram mapping, constructing a sequence processing network by using a U-Net model as a basis, integrating the space-time attention mechanism into a picking up model, using a deep coding feature fusion mechanism to supplement missing feature information, adjusting model parameters according to loss values and evaluation indexes to complete final construction of the model, and inputting the seismic signal data to be identified into a microseismic facies picking up model to obtain a seismic facies in-time picking up marking result. For example, the publication number is CN112799128A, and a method for detecting the earthquake signal and extracting the earthquake phase is disclosed, which is used for an earthquake detection system based on edge equipment. The edge equipment is realized by Jetson Nano chips, a lightweight deep learning model LCANet is constructed and arranged on the edge equipment, the seismic waveform data acquired by the equipment end are input into the edge equipment, and the seismic signal time sequence, the longitudinal wave seismic phase and the transverse wave seismic phase are output in real time. The LCANet model extracts the characteristic vector sequence describing the internal physical meaning of the seismic data from the input seismic waveform data through an encoder based on a reverse bottleneck residual block, acquires the characteristic vector sequence of the context information of the concerned time sequence under three tasks through a context awareness and attention module, and finally maps the characteristic vector to the characteristic space of the corresponding task through a mult