Reverse Time Migration Compensation Method Of Seismic Wave Based On Deep Learning

The viscoelastic strong attenuation geological structure will bring the amplitude weakening and phase distortion of the seismic wave,and the conventional reverse time migration compensation method is difficult to obtain high imaging resolution.This paper focuses on the deep learning method to study the compensation of seismic wave reverse time migration.Combined with the core idea of geophysical inversion,a seismic wave reverse time migration compensation method based on deep learning is proposed and implemented,which improves the imaging resolution of seismic data.First of all,this paper systematically summarizes the factors and parameters that affect the attenuation of seismic waves,and deeply explores its physical mechanism,and lists several methods for calculating the quality factor Q.For deep learning network architectures,this paper proposes two compensation models for different data types.Aiming at the attenuation compensation of seismic data in the form of two-dimensional wavefield images,a reverse time migration compensation method of seismic waves based on Cycle GAN is proposed.The model combines the Cycle GAN algorithm with the attention mechanism,which reduces the redundancy and waste of resource computing power and reduces overcompensation.In the experiment,the loss function is improved to a hybrid loss function of the cross-entropy loss function and the perceptual loss function,which makes it more suitable for the compensation of seismic waves in strongly attenuated geology.Aiming at the characteristics of time-domain dependence of the context of seismic data in the same block,a 3W-based convolutional block attention module(3W-CBAM)is proposed and applied to model construction.Compared with ordinary attention,the3W-CBAM structure The mechanism realizes the upgrade of dimensions and improves the efficiency of attention factor extraction in key areas.The results verify that the seismic wave reverse time migration compensation model is more ideal in terms of waveform structure similarity after incorporating the attention mechanism,and has certain application value for the attenuation processing of seismic data.Aiming at the attenuation compensation of seismic data in one-dimensional data format extracted from seismic curves,this paper proposes a seismic wave reverse-time migration compensation method based on DBNs,and introduces a gray correlation algorithm to design a complete attenuation compensation model based on DBNs+GRA,with the highest compensation accuracy.The rate can reach 95.38%,which verifies the feasibility and superiority of this method.According to the different characteristics of different attenuation distributions of the four types of sites,the weighted algorithm and the grey correlation algorithm are used to comprehensively evaluate the seismic wave compensation effect.Experiments verify the validity and applicability of the model.In the future,the model can be extended and applied to different geological exploration environments,which has broad application potential.In summary,the two deep learning-based seismic wave reverse-time migration compensation methods proposed in this paper have improved in robustness,transferability and accuracy,indicating the feasibility of seismic data processing and interpretation.Has wide application potential.