Research On Seismic Wave Feature Extraction Method Based On Auto-Encoder

Interpreting underground geological problems through seismic reflection data is one of the important means of exploration and development.The lateral changes of seismic reflection waves reflect the lateral changes of underground stratum characteristics to a certain extent.Therefore,it is of great significance to obtain underground structures and depict geological bodies by using seismic attributes and seismic wave group information in seismic facies analysis.This paper uses the powerful nonlinear feature extraction of deep learning technology and the ability of more abstract high-level features to represent categories or features of low-level features to extract seismic wave features,and implements a Sparse Denoising Auto-Encoder network model and clustering analysis method.The waveform is classified and applied to the data of the F3 area to help geological interpreters better identify the characteristics of the geological body.Related work is as follows:(1)Performance comparison of various autoencoders.Based on the research on the basic principles of various autoencoders,the focus is on in-depth research on Sparse Denoising Auto-Encoder(SDAE).The training experiment results of different autoencoders on the MNIST handwritten digit set show that SDAE can not only reduce noise and enhance the robustness of the algorithm;but the addition of sparse terms optimizes the network structure and can better extract the deep features of the data.(2)Designing and implementation of seismic wave feature extraction model based on stacked SDAE.Using the deep learning framework Tensor Flow,a deep stacked sparse denoising autoencoder model is constructed.Input seismic data,and extract more abstract features layer by layer through a four-layer sparse denoising autoencoder.The autoencoder of each layer takes the restoration of the input features as the optimization objective to learn model parameters.Through a large number of experiments,the optimal number of hidden layers in the encoding and decoding stages,the number of hidden layer nodes and the network connection weight parameters of SDAE are determined,and the effectiveness of the model is verified on artificial synthetic data and real seismic data.(3)The SDAE method is used to realize the cluster analysis of seismic wave characteristics of the F3 area data.Using different clustering algorithms to perform cluster analysis on the original seismic data,the fuzzy c-means clustering algorithm is optimized.By comparing the features of geological bodies based on the clustering results of SDAE feature extraction with the results of other methods(direct cluster analysis,cluster analysis based on principal component analysis feature extraction,cluster analysis based on Opend Tect software waveform classification),It is verified that the SDAE feature extraction method proposed in this paper has a better effect on waveform classification and recognition of geological features.