The laminar sedimentary structures of saline lacustrine mixed rocks affect both organic matter enrichment and reservoir storage performance. However, due to the small-scale nature of laminae, large-scale identification using well-logging data during reservoir exploration and development remains challenging. It is necessary to introduce a research method to identify and characterize the development of different types of laminae. Based on analyses of typical cores, XRD data, and well-logging curves from the upper member of the Xiaganchaigou Formation in the Yingxi area, five main types of laminae and six lamina combinations were classified. A Transformer-based intelligent recognition method was then applied to identify these lamina combinations from well-log data, with the Random Forest algorithm used as a comparative benchmark. Verification results show that the Transformer model achieves a higher total accuracy of 84% in lamina combination recognition. This study proposes a new approach for the conventional well-log characterization of laminae, in which the classification is established from the perspective of laminae genesis. It reflects the development patterns of lamina combinations driven by paleoenvironmental changes, and selects an appropriate intelligent recognition method to address the challenges in well-log characterization of such reservoirs. In terms of engineering applications, this study can accurately indicate the positions of high-quality reservoirs within sedimentary cycles during field development. It provides a sedimentary facies-controlled basis for the three-dimensional characterization of reservoir quality, thereby offering a valuable reference for the exploration and development of reservoirs formed under similar sedimentary conditions.
Fu et al. (Wed,) studied this question.