Research On Seismic Signal Feature Extraction And Cluster Analysis Method Based On Convolutional Neural Network

Nowadays,with the vigorous development of geophysical exploration and development technology,waveform classification technology as an important method for underground structural interpretation and reservoir prediction has attracted widespread attention from interpreters.At the same time,the complex engineering geological environment of oil and gas in China has also put forward higher requirements for exploration accuracy.Unlike post-stack seismic data,pre-stack seismic data is large in magnitude and high in dimension,and it also contains richer information about underground stratum structure.Using traditional processing methods can easily cause dimensional disasters and cause inaccurate classification results.In addition,due to the limited and inaccurate seismic data,uncertainties and multiple solutions still exist in the application of seismic attributes.Aiming at the above problems,this paper proposes an unsupervised clustering method of prestack seismic data based on convolutional neural network combined with deep learning algorithms.It mainly focuses on four aspects of prestack seismic data preprocessing,data reduction,feature selection and pattern recognition :(1)With the continuous deepening of artificial intelligence,convolutional neural networks have made many breakthroughs in the areas of image analysis,processing,and target detection.This paper introduces convolutional neural networks to reduce prestack seismic data by virtue of its powerful feature compression and extraction capabilities Dimension and extract its deep features.Because the weights are shared among the nodes in the convolutional neural network,the number of parameters is greatly reduced,the operation efficiency is improved,and it is suitable for processing prestack seismic data.(2)At present,there are more and more seismic attribute parameters that can be extracted and combined from prestack seismic data.The agglomeration hierarchical cluster analysis of the deep features of multi-attribute prestack seismic data can reduce the multiplicity of seismic interpretation to a certain extent.(3)In this paper,the feasibility of the algorithm is verified by identifying the sand body distribution in the seismic physical model,and it is applied in the actual work area.Compared with the PCA-based unsupervised clustering method,the lithological boundary is clearly described and the classification results are more accurate.In the end of this paper,the empirical summary of the hyperparameter settings and feature map selection principles in the deep convolutional autoencoder is summarized.