Construction Of A Pseudo Sample Library With Forward Record Constraints

In the field of oil and gas exploration,in order to accurately find deep concealed oil and gas reservoirs with complex conditions,a large amount of seismic data is often needed to help people do seismic interpretation.If the deep learning can be used to expand the relevant seismic data based on the existing seismic data,it can alleviate the problems of small number of samples and incomplete sample types for intelligent prediction and identification of sweet reservoirs in complex oil and gas reservoirs to some extent.Based on this,this paper conducts research on seismic data expansion methods based on data-driven methods,and two improved network structures: the improved WGAN-GP and the VAE-CFWGAN,are proposed for seismic data expansion to realize the construction of a pseudo sample database.According to experiments,the proposed method is effective.The main research content and results of this article are as follows:1.The method of using CFWGAN to expand seismic data under the constraint of forward recording is proposed.Aiming at the problem of the unsatisfactory effect of traditional WGAN-GP seismic data generated under forward record constraints,this paper improves and optimizes WGAN-GP,and proposes an improved network:CFWGAN.Through training and verification on the theoretical and actual data sets,it is proved that it is effective to use the CFWGAN to expand seismic data to realize the construction of a pseudo sample database under the constraints of forward records.2.The feasibility of using variational autoencoders for seismic data expansion was studied.Based on the related theoretical basis of variational autoencoders,this paper uses theoretical data to simulate and analyze the feasibility of applying variational autoencoders to seismic data expansion.Experiments show that the ability of the variational autoencoder network to extract and learn data features is very strong,if we can solve the problem that the data generated by the variational autoencoder has a certain degree of distortion,it is feasible to apply the variational autoencoder to the expansion of seismic data.3.The method of using the VAE-CFWGAN to expand seismic data under the constraint of forward recording is proposed.Although the CFWGAN can effectively generate seismic data,it also generates data out of the real data distribution,andalthough the variational autoencoder has a strong ability to extract and learn data features,the data it generates has a certain degree of distortion.This paper proposes another improved network structure: VAE-CFWGAN.By training and verifying on theoretical and actual data sets,it proves that the use of VAE-CFWGAN for seismic data expansion under the constraint of forward recording to achieve the construction of the pseudo sample library is effective,and the related results are better than CFWGAN.