Strike-slip fault identification is important to the exploration and development of fault-controlled fractured-vuggy car-bonate reservoirs,but the horizontal displacement of a strike-slip fault is ambiguous on seismic sections perpendicular to fault strike.Manual fault interpretation with heavy workload is greatly dependent on the experience of the interpreter.We propose an in-telligent method for strike-slip fault identification based on the deep residual network.The residual network is composed of three sub-networks for feature extraction,structure extraction,and denoising convolution,respectively.The sub-network of feature ex-traction is used to extract residual mapping features of seismic and fault prediction.The sub-network of denoising convolution is used to remove accumulated noises generated by the network.The sub-network of structure extraction is used to extract the residu-al mapping of boundary structure for fault interpretation.Multi-layer output fusion and transfer learning are adopted to avoid high-frequency loss in network-based prediction and enhance the robustness and generalization of fault classification and interpretation of different scales.Model tests using synthetic records show high accuracy,small missing rate,good continuity,distinct boundaries,and good anti-noise performance of identifying the faults with small displacement and strike slip on seismic sections with low sig-nal-to-noise ratio.The field data application to fault-controlled fractured-vuggy carbonate reservoirs in Fuman Oilfield,the Tarim Basin shows good results of identifying linear strike-slip faults,compressional torsional braided strike-slip faults,and extensional braided strike-slip faults.
国家科技期刊平台
NSTL
万方数据
