Application And Discussion Of Machine Learning In Faults Recognition And Seismic Facies Waveform Classification

With the continuous development and improvement of machine learning methods,a wave of artificial intelligence led by deep learning has emerged.The deep learning method represented by convolutional neural network(convolutional neural network)has made great achievements in image recognition,semantic recognition and other hot artificial intelligence fields,and attracted the interest of researchers in the field of seismic exploration.The interpretation of seismic data is an important part of seismic exploration,and fault interpretation and seismic facies analysis are important parts for seismic data interpretation.At present,the application of machine learning algorithm in seismic data interpretation mainly includes automation and efficiency improvement,such as fault interpretation and seismic automatic analysis.By giving full play to the characteristics of "intelligence" of machine algorithm,it explores some effects and precision beyond manual interpretation,such as seismic attribute analysis and seismic inversion.Therefore machine learning algorithm can play an important role in fault interpretation and seismic facies analysis.This paper is based on the requirements of relevant scientific research projects.The purpose is to solve the detailed characterization of small-scale faults in buried mountains and seismic facies analysis under complex geological conditions.This paper mainly focuses on the applicability and application effect of deep learning methods in fault recognition and seismic facies analysis,which including the construction and training of various deep learning network models,the adjustment of parameters,the generation and effectiveness of label data,and the solution of some algorithms such as overfitting.This paper discusses how to set up a series of deep learning methods for different application purposes and how to analyze the effect of the application of these methods.Through this research,the understanding of the key issues in deep learning algorithm,such as network model construction and error back propagation mechanism,is improved.Based on python language and Tensor Flow framework,convolutional neural network(CNN)calculation for fault recognition and seismic phase waveform classification is realized.The up-sampling structure of the CNN network model is improved,and the network structure is simplified without reducing the accuracy of the recognition results.The convolution layer structure is improved by using the residual learning module.The loss function is improved by increasing the penalty coefficient,and the generalization ability of the deep learning network is improved.The application of practical data shows that the fault identification accuracy based on CNN algorithm is improved.By combining the characteristics and advantages of supervised and unsupervised learning,a method for waveform classification of supervised and unsupervised deep learning is established,which improves the reliability of seismic phase analysis in the application of practical data.