Research On Fault Recognition Method Based On Convolutional Neural Network

Fault is an important part of geological structure that directly controls structural traps and oil and gas reservoirs,and it will directly affect the effect of injection,production and development.Therefore,accurate fault identification is of great help for geological researchers to analyze and evaluate underground structures and oil and gas reservoirs.In the process of seismic data interpretation,the conventional fault interpretation method is to interpret the fault manually by the interpreter,which is very time-consuming and subjective.At present,with the rapid development of machine learning and deep learning,using machine learning and deep learning can automatically learn target features,and has the advantage of learning multi-level features.Many scholars of geology and geophysics have proposed intelligent fault interpretation methods based on machine learning and deep learning.In this paper,the problem of fault recognition in 3D seismic data is regarded as a semantic segmentation problem of binary image.By using the powerful feature learning and feature extraction ability of convolutional neural network in deep learning,the existing model methods are optimized and improved to improve the efficiency and accuracy of automatic fault recognition in 3D seismic data.The main achievements are as follows:(1)This paper uses a fault recognition method based on the FCRes Net.This method adds residual learning units on the basis of Fully Convolutional Networks,which greatly reduces the difficulty of network learning targets and makes deep neural network training more easy.The fault recognition experiment test and analysis of the FCRes Net verify that the FCRes Net method can effectively identify faults and has high accuracy.(2)In order to better capture the characteristic information of a larger area to identify the faults in the earthquake,while reducing the amount of calculation parameters,this paper also uses an automatic fault identification method based on the Pyramid Scene Parsing Network.This method introduces a multi-scale pyramid pooling module,which reduces the loss of contextual information between different sub-regions,improves the ability to obtain prior information of the global scene,and obtains more fault feature information.Experiments show that although the parameters of the Pyramid Scene Parsing Network are less than that of the FCRes Net,and the training time is less than that of the FCRes Net,the fault recognition effect is better than that of the FCRes Net,in order to reflect the effectiveness of the method used in this paper.Compared with the existing excellent methods,the experimental results show that the method used in this paper has higher accuracy and good performance in noise resistance.Among them,the Pyramid Scene Parsing Network method has a better fault recognition effect.(3)Applied to actual 3D seismic data.The FCRes Net method and the Pyramid Scene Parsing Network method used in this paper are applied to actual 3D seismic data for fault identification,and compared with the existing excellent methods FCN-8s and U-Net,the results show that this paper uses the methods improve the ability to suppress noise interference and the continuity of the fault,especially the Pyramid Scene Parsing Network method is very good at suppressing noise interference.