Research On Automatic Fracture Recognition Method Of Volcanic Reservoir In CH47 Area

Fractures are well developed in volcanic reservoirs in Che47 well area,which plays an important role in oil and gas reservoir,migration and production.Because of the limited logging data in the study area,the corresponding blurred logging of fractures in conventional logging,the influence of lithology,fluid and other factors leads to the low accuracy of fracture identification,even difficult to identify;the corresponding characteristics of fractures in imaging logging are obvious,but the imaging processing process is cumbersome and the workload of manual identification and pickup is large,so it is necessary to propose new methods to identify fractures.After calibrating the conventional logging with the fracture identified in the previous step and the core and thin section data,Deep Labv3+,the most advanced image semantics segmentation model in the field of computer vision,is used to segment the crack area pixels in this paper.Based on the segmentation results,Hough transform is used to extract the crack shape parameters.Compared with the traditional method,this method can extract the fracture area more accurately and quickly,the MIo U improved to 83.2%,and obtain the fracture occurrence.In this paper,a new data-driven fracture identification method is proposed,which applies auto DL technology to conventional logging fracture identification.This method combines change point detection,sparse autoencoder and convolution neural network(CNN);change point detection layers logging data according to the statistical characteristics of relevant logging curves;sparse self-encoder sampled logging data from each layer into two-dimensional coding map of convolution layer,and extracts the first feature;convolution layer of convolution neural network extracts relevant fracture occurrence from two-dimensional coding map.This method has improve the accuracy from 84% to 91% compare to maching learning method.Finally,using the fracture prediction results of imaging logging and conventional logging,combined with the prediction of seismic attributes,the law of fracture development in the study area is analyzed,and it is clear that the fracture development in the study area is mainly controlled by volcanic lithofacies and faults.