Pore Throat Feature Extraction And 3D Structure Reconstruction Of Micro-CT Digital Core Based On Res-Unet

Petroleum energy is the core raw material for the industrial development of modern society.However,conventional oil and gas reservoir resources are limited,and the exploitation of unconventional reservoirs such as tight sandstone reservoirs is difficult due to the characteristics of low porosity,low permeability and strong heterogeneity.Therefore,the key to reservoir reconstruction and stimulation is to construct the core pore throat structure accurately and comprehensively.In view of the above situation,this paper takes the tight sandstone gas reservoir of Xiashihei Formation in Area J of Ordos Basin as the research object,and proposes a set of methods from two-dimensional slice extraction of pore-throat features to threedimensional reconstruction of pore-throat structure.The main research results are as follows:(1)In view of the situation that there is no or only one guiding standard for the division of pore-throat structure in the traditional method,the nuclear magnetic threshold segmentation method is proposed to divide the basic pore-throat structure by combining Micro-CT digital core technology,nuclear magnetic resonance technology and core processing software Avizo.Threedimensional digital core and its basic pore-throat structure were sliced to obtain twodimensional core slices and corresponding binarized labels of pore-throat features.Label slices that need to be manually supplemented were screened for supplementary labeling and used as deep learning data sets.(2)The Res-Unet model is improved to improve the feature extraction ability of the model,considering that the pore throat structure of tight sandstone contains a large number of microfractures and other fine features of veins.The Miou value of the connected pore-throat features of the test core slices segmented by the Res-Unet model was up to 84%,and that of the corresponding total pore-throat features was up to 81%,which was higher than the extraction results of other deep learning models and the nuclear magnetic threshold segmentation method.The model test results showed that the model could learn part of the pore-throat feature information that needed to be manually marked.(3)The digital tight sandstone cores in the same area and well depth as the training cores were selected,the Res-Unet model was used to extract the slicing pore-throat characteristics,and the predicted pore-throat structure was obtained by sequential slicing superposition imaging method,and combined with the basic pore-throat structure extracted by the nuclear magnetic threshold segmentation method,to comprehensively complete the three-dimensional structure reconstruction of tight sandstone pore-throat structure.The pore-throat parameters of the reconstructed pore-throat structure were quantitatively analyzed.The results show that the reconstructed pore-throat structure in this paper matches well with the test results and has high practical application value.