Intelligent identification of laminae and laminasets in the Qiongzhusi Formation shale， southern Sichuan Basin and their impacts on reservoirs

The efficient and high-precision quantitative characterization of laminae is critical to shale reservoir assessment. However， research on shale laminae faces some challenges， including the low accuracy and cumbersome nature of lamina identification methods， as well as the failure to establish the relationship between macroscale and microscale laminae. In this study， we investigate laminae at varying scales in the Qiongzhusi Formation within the southern Sichuan Basin based on thin section and core observations， imaging logs， as well as analytical and test data. Using image and signal processing， Light Gradient Boosting Machine （LightGBM）， convolutional neural networks （CNNs）， Long Short-Term Memory （LSTM） network， and Kolmogorov-Arnold Networks （KANs）， we develop methods to identify the laminae of varying scales and establish a laminaset identification and prediction model （Laminae-Net）. Furthermore， we develop a method for the unified characterization of laminae at varying scales and explore the impacts of lamina characteristics on shale reservoirs. The results indicate that the proposed thin section-based lamina identification method using image and signal processing allows for the accurate discrimination of bright and dark laminae， contributing to the precise characterization of lamina thickness and density. In contrast， the proposed lamina identification method based on FMI images can efficiently characterize the lamina number and thickness with high precision. Eight types of laminasets primarily occur in the Qiongzhusi Formation in the southern Sichuan Basin. Using the Laminae-Net model， these laminasets can be identified with an accuracy of up to 95.4% on the test set. These laminae preferentially occur in the 1st， 3rd， 5th， 6th， 7th， and 8th sublayers of the formation. ​Scale invariance of lamina development frequency​ is verified through integrated thin section and image logging analysis.. A first-of-its-kind cross-scale lamina calculation method is built for lamina characterization， achieving a goodness of fit （R2） of up to 0.813. The TOC content shows a negative correlation with the lamina density， while the porosity and horizontal permeability exhibit positive correlations with the lamina density. This study provides methods for identifying and analyzing laminae at varying scales， expanding the scope of lamina research while also serving as a reference for advancing the research.