Small‐Scale Strike‐Slip Fault Identification Method Based on Deep Learning

Intracratonic strike‐slip faults play a vital role in controlling the migration and accumulation of hydrocarbons in the ultra‐deep carbonate reservoirs of the Shunbei area, Tarim Basin. While the major fault zones define the structural framework, accurately characterizing the associated small‐scale strike‐slip faults remains a significant challenge due to their subtle seismic responses and low signal‐to‐noise ratios. Traditional geometric attributes often fail to preserve fault continuity, while conventional deep learning approaches are limited by the scarcity of labelled field data and the lack of specific samples for small‐scale structures. To address these issues, this paper presents a high‐precision fault detection method utilizing a deep neural network trained on a geologically constrained synthetic dataset. By integrating the specific stratigraphic framework and fault patterns of the Shunbei area, we construct a specialized training dataset via point spread‐function convolution, enabling the network to learn the kinematic features of small‐scale strike‐slip faults. Application to the Shunbei No. 6 fault zone demonstrates that the presented method significantly outperforms both traditional seismic attributes (e.g., coherence) and general deep learning models trained without specialized samples. The fault detection results exhibit superior vertical continuity, higher resolution and superior noise suppression. Validation with well logging data confirms a high consistency between the predicted faults and drilling evidence. Practically, the proposed method provides a reliable geophysical basis for identifying favourable reservoir zones and optimizing well deployment in ultra‐deep exploration.