Elastic wave multi-parameter dual-drive inversion method and system based on multi-task learning

CN117130054BActive Publication Date: 2026-07-24SHANDONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SHANDONG UNIV
Filing Date
2023-07-25
Publication Date
2026-07-24

AI Technical Summary

Technical Problem

Existing multi-parameter inversion methods for elastic waves suffer from problems such as heavy reliance on the initial model, easy getting trapped in local optima, and low computational efficiency. Deep learning methods have weak generalization ability in elastic wave data applications and are not designed for the characteristics of elastic wave data and elastic parameter models.

Method used

A multi-parameter dual-drive inversion method for elastic waves based on multi-task learning is adopted. Combining deep neural networks and elastic wave equations, the mapping relationship between elastic wave data and elastic parameters is established through multi-task learning and forward modeling networks. By combining the data mining capabilities of deep learning with physical laws, a decoder with a serial structure and skip connections are designed to optimize the network model.

Benefits of technology

It improves the accuracy of elastic parameter inversion, enables accurate imaging of underground media, enhances the prediction accuracy of wave velocity and geological structure, and solves the problems of local optima and insufficient generalization ability in traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117130054B_ABST
    Figure CN117130054B_ABST
Patent Text Reader

Abstract

The present application belongs to the field of geophysical exploration, and provides an elastic wave multi-parameter double-drive inversion method and system based on multi-task learning. The method comprises: obtaining a real elastic parameter model and corresponding real elastic wave data; wherein the real elastic parameter model comprises a P-wave velocity model, a S-wave velocity model and a density model; using an elastic wave multi-parameter inversion network based on multi-task learning to learn the mapping relationship between the real elastic wave data and the real elastic parameter model, and obtain a predicted elastic parameter model; using a forward calculation network to calculate the elastic wave data corresponding to the predicted elastic parameter model in combination with a determined elastic wave equation; calculating a joint loss function to optimize the elastic wave multi-parameter inversion network based on multi-task learning; and based on the optimized elastic wave multi-parameter inversion network based on multi-task learning, obtaining a P-wave velocity model, a S-wave velocity model and a density model graph of the underground medium according to the obtained elastic wave data, and realizing seismic data inversion.
Need to check novelty before this filing date? Find Prior Art