CN-122017974-A - Seismic facies pickup method and system based on meta knowledge memory base enhancement

CN122017974ACN 122017974 ACN122017974 ACN 122017974ACN-122017974-A

Abstract

The invention belongs to the field of seismic facies pickup, and discloses a seismic facies pickup method and system based on the enhancement of a meta knowledge memory base, wherein the method comprises the steps of obtaining seismic waveform data; the method comprises the steps of processing seismic waveform data through a seismic phase pickup basic model encoder to obtain basic time sequence characteristics, generating an index through regional characteristic encoding, searching in a meta knowledge memory base to obtain a memory vector, fusing the basic time sequence characteristics and the memory vector by using a memory enhancement module to obtain enhanced characteristic representation, decoding the enhanced characteristic representation by using a decoder to output a P wave probability sequence and an S wave probability sequence, and completing a seismic phase pickup task. The invention solves the problems that the traditional fine tuning method is forgotten and knowledge of similar areas cannot be reused, is particularly suitable for continuous construction scenes of the earthquake monitoring network, can continuously accumulate and utilize knowledge along with continuous addition of new stations, and provides more and more intelligent earthquake phase pickup capacity for an earthquake early warning system.

Inventors

ZHU YADONGYANG
ZHAO XIAOMIN
XU JINGYU
Yu Lanya

Assignees

北京石油化工学院

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (10)

1. A seismic facies pickup method based on meta knowledge memory base enhancement, the method comprising: acquiring seismic waveform data; Processing the seismic waveform data through a seismic phase pickup basic model encoder to obtain basic time sequence characteristics; Constructing a meta knowledge memory base; generating an index through regional feature codes, and searching in a meta knowledge memory base to obtain a memory vector; fusing the basic time sequence characteristics and the memory vectors by using a memory enhancement module to obtain enhanced characteristic representation; and decoding the enhanced characteristic representation by using a decoder to output a P-wave probability sequence and an S-wave probability sequence, so as to complete the vibration phase pickup task.
2. The method of claim 1, wherein the method of building a metaknowledge repository comprises: And learning and fusing adjacent information according to the relation diagrams of different areas by using the graphic neural network to enhance the representation of the seismic area characteristics and construct a meta knowledge memory base.
3. The method of claim 2, wherein the indexing is generated by region feature encoding, and the retrieval is performed in a meta knowledge memory base to obtain the memory vector comprises: ; ; Wherein, the 、、 Representing the weights of the knowledge of the memory cells in the fusion, Representing the retrieved historical meta-knowledge relating to the characteristics of the target area, The sharpness is indicated by the fact that, Representing the memory vector, fusing the retrieved meta-knowledge, In order to enhance the representation of the region features, In order to weight the attention coefficient, As a feature of the region it is, For a nonlinear activation function, i and j represent different regions, The number of support set samples for region i is indicated.
4. A method according to claim 3, wherein the fusing of the base timing features and the memory vectors using a memory enhancement module results in an enhanced feature representation comprising: ; ; Wherein, the Representing the mechanism of the gating, In order to modulate the signal, Is a time sequence feature.
5. The method of claim 4, wherein decoding the enhanced representation of the features using a decoder to output a P-wave, S-wave probability sequence, the method of performing a seismology phase pickup task comprising: ; ; Wherein, the Representing the wave type, hup represents the waveform sequence obtained after upsampling, Representing a full connection layer calculation of the full connection layer, The probability sequences of P-waves and S-waves are represented, and l represents the number of layers.
6. A seismic facies pickup system based on the enhancement of a meta knowledge memory base, which is used for realizing the method of any one of claims 1-5, and is characterized in that the system comprises an acquisition module, a coding module, a construction module, a retrieval module, a fusion module and a decoding module; the acquisition module is used for acquiring the seismic waveform data; the coding module is used for processing the seismic waveform data through a seismic phase pickup basic model coder to obtain basic time sequence characteristics; the construction module is used for constructing a meta knowledge memory base; the retrieval module is used for generating an index through regional feature codes, retrieving in a meta knowledge memory base and obtaining a memory vector; The fusion module is used for fusing the basic time sequence characteristics and the memory vectors by using the memory enhancement module to obtain enhanced characteristic representation; and the decoding module is used for decoding the enhanced characteristic representation by using a decoder to output a P-wave probability sequence and an S-wave probability sequence, so as to complete the vibration phase pickup task.
7. The system of claim 6, wherein the process of building a metaknowledge repository comprises: And learning and fusing adjacent information according to the relation diagrams of different areas by using the graphic neural network to enhance the representation of the seismic area characteristics and construct a meta knowledge memory base.
8. The system of claim 7, wherein the indexing by region feature encoding, retrieving in the meta knowledge memory base, and deriving the memory vector comprises: ; ; Wherein, the 、、 Representing the weights of the knowledge of the memory cells in the fusion, Representing the retrieved historical meta-knowledge relating to the characteristics of the target area, The sharpness is indicated by the fact that, Representing the memory vector, fusing the retrieved meta-knowledge, In order to enhance the representation of the region features, In order to weight the attention coefficient, As a feature of the region it is, For a nonlinear activation function, i and j represent different regions, The number of support set samples for region i is indicated.
9. The system of claim 8, wherein fusing the base timing features and the memory vectors using a memory enhancement module to obtain an enhanced feature representation comprises: ; ; Wherein, the Representing the mechanism of the gating, In order to modulate the signal, Is a time sequence feature.
10. The system of claim 9, wherein decoding the enhanced representation of the features using a decoder to output a P-wave, S-wave probability sequence, the process of performing a seismology phase pickup task comprising: ; ; Wherein, the Representing the wave type, hup represents the waveform sequence obtained after upsampling, Representing a full connection layer calculation of the full connection layer, The probability sequences of P-waves and S-waves are represented, and l represents the number of layers.

Description

Seismic facies pickup method and system based on meta knowledge memory base enhancement Technical Field The invention belongs to the field of seismic facies pickup, and particularly relates to a seismic facies pickup method and system based on metadata knowledge memory base enhancement. Background The deep learning technology has advantages in processing a large amount of complex data by virtue of the strong feature extraction capability, and has been widely applied to seismic facies pickup tasks. However, existing deep learning models still face the following challenges: In order to adapt to new regional data, the model needs to be finely adjusted, knowledge features learned in the original region are often forgotten, so that the seismic phase pickup performance in the original region is obviously reduced, the model is difficult to serve a plurality of stations in different regions at the same time, and the deployment efficiency of the deep learning method in large-scale seismic monitoring is severely limited. Meanwhile, the geological structure types, the seismic source mechanisms and the waveform propagation characteristics among different seismic areas may have inherent similarity, but the existing method lacks an effective knowledge multiplexing mechanism, and cannot fully mine and utilize the trans-regional common characteristics, so that the waste of data resources and learning experience is caused. In addition, the continuous learning capacity of the model is insufficient, the traditional method adopts a static training paradigm, and when new region annotation data is acquired, the whole model needs to be retrained, so that the calculation cost is high, and the gradual accumulation and the knowledge migration of knowledge cannot be realized. In recent years, meta-knowledge has shown significant potential in complex data reasoning. By constructing a sustainable evolution meta knowledge memory base, the efficient accumulation, retrieval and multiplexing of cross-region seismic knowledge can be realized, and a new thought is provided for seismic phase pickup. Disclosure of Invention The invention provides a seismic facies picking method and system based on the enhancement of a meta knowledge memory base, which solve the problems that the traditional fine tuning method is forgotten and knowledge of similar areas cannot be reused, are particularly suitable for continuous construction scenes of a seismic monitoring network, can continuously accumulate and utilize knowledge along with the continuous addition of a new station, and provide more and more intelligent seismic facies picking capability for a seismic early warning system. In order to achieve the above object, the present invention provides the following solutions: a seismic facies pickup method based on meta knowledge memory base enhancement, the method comprising: acquiring seismic waveform data; Processing the seismic waveform data through a seismic phase pickup basic model encoder to obtain basic time sequence characteristics; Constructing a meta knowledge memory base; generating an index through regional feature codes, and searching in a meta knowledge memory base to obtain a memory vector; fusing the basic time sequence characteristics and the memory vectors by using a memory enhancement module to obtain enhanced characteristic representation; and decoding the enhanced characteristic representation by using a decoder to output a P-wave probability sequence and an S-wave probability sequence, so as to complete the vibration phase pickup task. Preferably, the method for constructing the meta knowledge memory base comprises the following steps: And learning and fusing adjacent information according to the relation diagrams of different areas by using the graphic neural network to enhance the representation of the seismic area characteristics and construct a meta knowledge memory base. Preferably, the method for generating the index through the regional feature code and searching in the meta knowledge memory base to obtain the memory vector comprises the following steps: ; ; Wherein, the 、、Representing the weights of the knowledge of the memory cells in the fusion,Representing the retrieved historical meta-knowledge relating to the characteristics of the target area,The sharpness is indicated by the fact that,Representing the memory vector, fusing the retrieved meta-knowledge,In order to enhance the representation of the region features,In order to weight the attention coefficient,As a feature of the region it is,For a nonlinear activation function, i and j represent different regions,The number of support set samples for region i is indicated. Preferably, the method for fusing the basic time sequence feature and the memory vector by using a memory enhancement module to obtain the enhanced feature representation includes: ; ; Wherein, the Representing the mechanism of the gating,In order to modulate the signal,Is a time sequence feature. Prefe