Velocity field modeling method based on deep learning
Seismic velocity field modeling plays a crucial role in oil and gas exploration and development.With the advancement of neural networks,deep learning has been widely applied to velocity field modeling due to its powerful nonlinear mapping capabilities.However,the mapping relationship between velocity models and seismic data is complex,and training using common shot gathers often leads to unstable network training and poor generalization ability.To address this issue,this paper proposes a method based on the U-Net++neural network for layer velocity model through the Common Midpoint Gather(CMP)and stacked velocity spectrum.Building upon the existing the common shot gather and stacked velocity spectrum mapping layer velocities,this method further leverages the one-to-one correspondence between CMP and stacked velocity spectrum,harnessing the powerful feature extraction and restoration capabilities of U-Net++.By learning the crucial features from seismic data,it generates precise subsurface velocity models.Experimental results on the test dataset demonstrate that the feature combination of common midpoint gather and stacked velocity spectrum exhibits superior stability and noise resistance compared to the feature combination of shot gather and stacked velocity spectrum.Constructing layer velocity models using deep learning enables efficient prediction of velocity models with similar subsurface structures through network training.This significantly improves the efficiency of building layer velocity models,making the integration of geophysics and deep learning promising for practical seismic exploration.
万方数据
维普
