Abstract: Based on the seismic，logging，core and dynamic data，the internal architecture of fault-controlled carbonate reservoirs in Shunbei No. 1 fault zone of Tarim Basin was divided. Based on the hierarchical division， a 3D geological model was established through seismic data attribute extraction and conversion，deep learning， target-based schematic point process simulation，and discrete fracture network simulation（DFN）. Numerical simulation of oil and gas reserves and reservoirs was carried out with the model，and the fitting results were compared with the actual production data. The results show that：（1）The Ordovician fault-controlled reservoirs in the study area were categorized into five levels，from large to small，including strike-slip fault-influenced zones， fault-controlled bodies，cave-like，intra-cave-like cluster-filling and fracture zones.（2）The strike-slip faultinfluenced zones has segmentation due to stress differences，which can be subdivided into extrusion section， pullout section and translation section. The fault-control bodies develop six kinds of planar combination styles， including fault intersection type and single-branch slip type in the pullout section，double-fault staggered type and double-fault intersection type in the translation section，and double-fault twisted type and double-fault intersection type in the extrusion section. The cave-like is characterized by bead-like reflections on the seismic section. The intra-cave-like is divided into fenestration and inter-fenestration（bedrock），and the fenestration can be further divided into clusters（breccia zones）and inter-clusters（fracture zones），and the whole is characterized by fenestration structure，with better physical properties of the clusters. The fracture zones are the main reservoir space for cave-like structures，which are more developed within clusters than between clusters，and more developed in Yijianfang Formation than in Yingshan Formation. High-angle fractures are mainly developed in Yijianfang Formation and Yingshan Formation，while horizontal fractures are mainly developed at the joints between the two.（3）The error between the oil and gas reserves predicted by the geological model and the geological analysis reserves is 1.75%. The simulated production well formation pressure and cumulative liquid production results of the model are highly consistent with production performance，with a fitting error of less than 10%.

Key words: strike-slip fault zone, fault-controlled bodies, cave-like, carbonate reservoir, 3D geological modeling, Deep learning, Ordovician, Shunbei No. 1 fault zone, Tarim Basin