Research On Seismic Facies Recognition Based On Generative Adversarial Network And Its Application

With the advent of the era of big data,deep learning utilizes a multi-level model structure to extract multi-level feature with different complexity extracted from complex data,such as images,automatically through the back propagation algorithm(BP algorithm).This end-to-end learning mode is widely used in various fields.Seismic lithology recognition is an essential part of reservoir parameter prediction.In the process of data acquisition,the labeled data account for only a small portion due to high drilling cost,and it is difficult to achieve the data size required for deep learning training,resulting in a significant variance of the recognition model.For this shortage,in this thesis,a semi-supervised algorithm based on Generative Adversarial Network(GAN)with Gini-regularization is proposed,called SGAN＿G.Adding Gini-regularization term to the loss function can improve the speed of convergence and generalization ability of the model,which is proved theoretically and verified by comparative experiments.Then SGAN＿G is applied to the lithology identification field,which has significantly improvement compared to previous recognition models.Due to the local correlation of seismic data,we use the sampling method with multiple-sampling points of seismic data as input,and implicitly use the formation information to further improve the seismic lithology recognition results.