Extraction And Reconstruction Of Fault Surface Based On Convolutional Neural Network

In geological exploration,a key part is the interpretation of seismic data,and fault interpretation is the core and foundation in seismic data interpretation.The interpretation of faults has a very important impact on the development and structure of the crust and the extraction and distribution of natural gas and petroleum.Initially,geological researchers used manual methods to explain faults and manually interpreted the horizons of three-dimensional volumetric data.However,such an interpretation method has the disadvantages of being time consuming,difficult and impossible to repeat,and the interpretation process is complicated and the error is obvious.Then a series of methods for calculating faults using the method of calculating attributes have been proposed.However,the selection and calculation of seismic attributes are often complicated.With the rapid development of artificial intelligence and convolutional neural network technology,the use of convolutional neural network technology for automatic fault identification makes it possible to improve the efficiency and accuracy of 3D fault identification under massive seismic data,and thus becomes a very The key research direction.In this thesis,based on the 3D seismic amplitude data,the method of fault surface extraction and reconstruction based on convolutional neural network is studied and proposed.It is mainly divided into fault identification and fault surface extraction and reconstruction.The main work and innovation of this article are as follows:1.Aimed at a large number of manual interpretations of difficult faults and complex calculation of seismic attributes,this thesis proposes a fault identification method based on convolutional auto-encoder neural network.The convolutional self-encoder neural network is used to automatically learn features from a large number of seismic amplitude volume data,avoiding the manual selection and calculation of attribute feature uncertainty.Using the advantage of unsupervised learning seismic amplitude data,the convolutional self-encoder neural network has a better fault recognition effect compared to traditional convolutional neural network fault identification methods.2.In order to achieve faster and more accurate fault identification,this thesis proposes a method based on full convolutional network for fault recognition of 3D seismic amplitude data.Using the full convolutional network for fault identification can create a fast end-to-end fault semantic segmentation model,and can obtain more accurate fault recognition results.3.In order to realize the visualization of the fault surface,a fault surface extraction and reconstruction method is proposed in this thesis.This method first uses the skeleton extraction method combined with the distance field and refinement to obtain a three-dimensional fault point cloud that reflects the topology of the fault.Then,a fast and direct three-dimensional tomographic surface reconstruction method is used,so that the reconstructed fault surface not only has good continuity but also retains the complicated spatial topology of the fault itself.This thesis validates the method proposed in this thesis by using actual 3-D seismic amplitude data in Area B of Southwest China.The experimental results show that the method of fault surface extraction and reconstruction based on convolutional neural network proposed in this thesis has completed efficient and accurate fault identification and finally achieved rapid fault surface reconstruction which is continuous and conforms to its own spatial topology.