Grading evaluation of deep shale reservoirs based on Bayesian optimized LightGBM algorithm

Conventional approaches, including multivariate regression, empirical formulas, and petrophysical models, often fail to adequately capture the complex nonlinear relationships between logging curves and reservoir parameters, resulting in limited accuracy for grading prediction of shale reservoirs. This study focuses on deep shale reservoirs within the Wufeng Formation (O 3 w ) and the first submember of the first member of the Longmaxi Formation (S 1 l 1 1 ) in the western Chongqing area. A shale reservoir classification model based on a Bayesian-optimized LightGBM algorithm was developed, and the SHAP algorithm was employed to quantitatively evaluate the contribution of individual logging curves. Subsequently, the proposed model was applied to the grading evaluation of deep shale reservoirs within the target area. The results demonstrate that, compared with the regression-based strategy, the classification-based strategy achieved significant improvements in model complexity, computational efficiency, and reservoir identification performance. A classification model of shale reservoir was established using the classification scheme, yielding weighted precision and weighted recall values of 89.7% and 89.6%, respectively, for the LightGBM algorithm. These values are superior to those obtained using the RF algorithm (87.52% and 86.96%, respectively) and the SVM (83.61% and 81.8%, respectively), on the testing dataset. DEN, GR, and CNL logs were identified as the most critical logging curves for identifying Type I and Type III shale reservoirs, whereas DEN, AC, and CNL logs exhibited heightened importance in identifying Type II reservoirs. Furthermore, the logging curves exerted complex nonlinear influence on the model decision-making process. The grading evaluation results show that Type I reservoirs are predominantly developed in the upper O 3 w and the first sublayer of S 1 l 1 1 . Overall, the Bayesian-optimized LightGBM algorithm facilitates efficient and accurate identification of shale reservoir types, thereby offering a novel approach for shale reservoir classification and evaluation.