Research On Prestack Seismic Waveform Classification Method

In the field of oil and gas exploration,seismic facies maps are generally classified and generated by using pattern recognition technology.It has an important role in the determination of underground oil and gas reservoirs.Prestack seismic waveforms have higher dimensions,greater data volume,and richer information than poststack seismic waveforms.Therefore,this paper mainly studies the use of prestack seismic waveform classification techniques to generate better seismic facies maps.Prestack seismic signals have high dimensions,and direct processing can easily lead to dimensional disasters,excessive computation,and inaccurate classification effects.The existing supervised learning method uses well logging information as tagged data to train the model,and then uses the model to perform seismic phase classification.The number of logging information is very less in large-scale seismic data,which can easily lead to overfitting and other problems,making the classification result inaccurate.In order to solve the above problems,this paper combines deep prestack seismic preprocessing,dimension reduction and feature extraction,clustering algorithms,and semi-supervised learning with deep learning.The specific work is as follows:First,propose a prestack seismic waveform classification method based on deep convolutional autoencoder:(1)Aiming at the characteristics of high dimensionality and complex information of prestack seismic waveforms,a deep convolutional autoencoder is introduced to extract features of prestack seismic waveforms.The deep convolutional autoencoder has very good feature extraction capability and can extract the deep nonlinear features of prestack seismic signals.At the same time,because the weights of the convolutional neural network are shared,the computational complexity of this algorithm is very low,and it is suitable for large-scale prestack seismic waveform data.(2)Introduce a fuzzy self-organizing map network algorithm to classify the features of prestack seismic waveform data.In the traditional unsupervised classification method,the classification result of each sample data is determined,that is,the samples are assigned to a specific cluster.The algorithm uses a membership degree to represent the classification result of each sample,and thus uses this membership degree to display the results of the seismic phase classification.The seismic facies map generated by using the fuzzy self-organizing map network is richer than the information provided by the traditional methods,the error situation is lower,and it can reflect the geological structure more,providing a basis for further seismic interpretation.Second,propose a semi-supervised prestack seismic waveform classification method based on deep convolution generation network:In general,large-scale seismic data has only a small amount of well logging data(ie,tagged data),which is sparse relative to the entire amount of data.If a supervised classification method(such as SVM,neural network,etc.)is used,it will be very Overfitting which is easy to produce.To solve this problem,this paper proposes a semi-supervised prestack seismic waveform classification method based on deep convolutional generated adversarial network.This method not only retains the feature extraction ability of deep convolutional neural network,but also can assist training with tagged data.The results generated using this algorithm are more precise and reasonable.