Abstract

Summary The low-order faults increase the complexity of the fault-block oilfield. The identification and study of low-order faults are crucial for enhancing the oil recovery. In this study, we propose a low-order faults identification method of carbonate reservoir based on 3D attention-based U-Net. The proposed method initiates by training a primary fault identification model using synthetic data and real seismic data with hand-marked labels, considering it as a pre-trained model. Subsequently, ant-tracking seismic attribute data is utilized as labels, and transfer learning is applied to obtain a lower-order fault identification model based on the pre-trained model. The identification results on 3D real seismic data indicate that the method exhibits potential for identifying low-order faults in carbonate reservoirs.