Research On Sedimentary Microfacies Recognition Based On Deep Learning

Sedimentary facies is a stratigraphic unit with certain lithologic and paleontological characteristics,which can reflect the formation environment and geological changes of sediments under certain natural environment.The productivity distribution of oil and gas reservoirs is closely related to sedimentary facies,especially the formation factors and spatial distribution rules of oil and gas reservoirs are controlled by sedimentary facies to some extent.Therefore,the study of sedimentary facies,especially sedimentary microfacies,has become the focus of oil and gas exploration and development.The traditional identification of sedimentary microfacies is usually through the analysis of core,rock thin section,scanning imaging and other logging data to identify the sedimentary microfacies.It also uses the combination of logging data and modern mathematics to identify the sedimentary microfacies quantitatively.However,the traditional sedimentary microfacies identification methods cannot avoid the risks caused by artificial parameter selection and complex operation,and their data processing ability is obviously insufficient.In view of the shortcomings of traditional methods for identification of sedimentary microfacies,this paper starts from two perspectives of logging data expansion and sedimentary microfacies identification.In order to overcome the problem of small amount of logging sample data,the logging data is firstly transformed into two-dimensional image form,and then the logging data is enhanced and expanded by using generated admittedly network Star Ga N.Secondly,Googlenet,RESNET-18 and improved Py Conv Res NET-18 residual network models were selected to identify sedimentary microfacies based on the morphological elements of sedimentary microfacies logging curve and the advantages of deep convolutional neural network in image processing.Experiments using four types of logging curve is used to model of training and testing,through the contrast analysis of three kinds of convolution model performance of neural network model and the results found that improved Py Conv Res Net-18 residual network model comprehensive performance is relatively good,sedimentary microfacies recognition effect is better,can be effectively used in the research of sedimentary microfacies identification,The results also prove the feasibility of the deep learning algorithm in the identification of reservoir sedimentary microfacies.