Lithology identification in volcanic reservoirs by using the Enhanced Kernel Fisher Discriminant: a case study in the eastern Junggar Basin, western China

The volcanic reservoirs of the Carboniferous Batamayineishan Formation in the eastern Junggar Basin exhibit significant lithological heterogeneity. Conventional well logs show complex responses, and core samples are scarce, making lithology identification challenging. To address this, an Enhanced Kernel Fisher Discriminant (EKFD) model was developed. The process involves two steps: first, integrating and clustering the step-like variations in well logs using Long Short-Term Memory (LSTM) networks and K-Means to create a Lithology Indication Curve (LIC). This LIC is then used to optimize the Kernel Fisher Discriminant (KFD) model, improving its ability to distinguish between lithologies. To validate the effectiveness of the EKFD in lithology identification, two cored sections not included in the training dataset were utilized. Compared with commonly used machine learning algorithms such as XGBoost and SVM, the EKFD exhibits superior performance in both lithology identification accuracy and consistency with core observations. SHAP analysis revealed that the density log is the most important feature for discrimination. The EKFD model proves highly effective for lithology identification in these complex volcanic reservoirs.