Intelligent characterization of ultra-deep carbonate strike-slip fault zones based on 3DResNeSt-UNet: a case study of the YueMan Block in the Tarim Basin

The strike-slip fault zones (SSFZ) in the ultra-deep carbonate rocks of the Tarim Basin provide crucial space for the migration and accumulation of oil and gas. Traditional 3D seismic interpretation methods have limitations in identifying SSFZ. This study proposes a seismic interpretation method based on the 3DResNeSt-UNet to identify SSFZ more effectively. This network integrates the 3DUNet and ResNeSt modules, using synthetic 3D seismic data and its corresponding label data as inputs for training. The synthetic seismic data incorporates geological knowledge and geological body parameters. Initially, by combining field outcrop observations, logging, and seismic data, the geological patterns of SSFZ are summarized. Guided by these geological patterns and based on the geological body parameters extracted from logging and seismic data, synthetic 3D seismic data and its corresponding label data are generated. The results indicate that the accuracy of the 3DResNeSt-UNet model on the training data exceeds 98%. The trained model achieves good recognition results on the seismic data of the YueMan block. Compared with traditional seismic interpretation results, the model's recognition accuracy is significantly improved and more aligned with geological understanding. Overall, the 3DResNeSt-UNet provides a new effective method for identifying SSFZ and has great potential for application in similar seismic interpretation scenarios.