庆城油田三叠系延长组长7段湖底扇储层三维智能建模OA

The sublacustrine fan depositional system in the Hua H100-He H60 block of Qingcheng Oilfield exhibits the characteristics of small-scale sand-bodies,complex lobe stacking patterns,and strong spatial heterogeneity.Therefore,it is difficult to accurately charac-terize the 3D distribution patterns of thin reservoirs using traditional modeling methods,which constrains the efficient development of tight oil reservoirs.A new modeling method based on generative adversarial networks(GANs)was proposed to improve development effect,re-constructing a 3D geological model from 2D cross-sections.The model's ability to extract geological structural features is effectively en-hanced by integrating the multivariate heterogeneous data of 2D geological profiles and geological plans and establishing a joint training framework.2D slicing of the generated 3D model is performed,and the 2D slices are compared with actual 2D sample sets in different di-rections using a multi-directional discriminator to construct a domain-adaptive adversarial loss function.The network parameters of the gen-erator is adjusted in real-time by iterative feedback to ensure the 3D geological model to conform to the geological characteristics of the ac-tual sample set,ultimately achieving accurate reconstruction of 3D geological body from 2D profiles.The application results show that the new method accurately establishes a 3D geological model of sublacustrine fan microfacies sand bodies,and achieves precise characteriza-tion of the spatial superposition relationship of gravity flow channel-lobe complex.The verification of 5 dilution wells shows that the pre-diction accuracy of microfacies is79.4%,indicating the high accuracy of the model.This research achievement enriches the theory of in-telligent modeling and provides effective technical support for intelligent prediction of tight oil reservoirs and horizontal well development.