Enhanced Total Organic Carbon Estimation in the Longmaxi Shale Formation: Integrating Unsupervised Clustering with a Stacked Hybrid Machine and Deep Learning Approach

The Longmaxi formation located within the Sichuan Basin has substantial heterogeneity with respect to its mineral constituents, organic materials, and deposition features. The highly complex internal structures and variations of the formation make the prediction of total organic carbon (TOC) very challenging. This prediction is essential for the evaluation of a shale reservoir. To address this issue, both machine learning (ML) and deep learning (DL) techniques have been applied to several well log data sets with the integration of both supervised and unsupervised learning approaches to enhance and enhance its geological interpretability. Supervised models employed training data consisting of the Uranium (U), bulk density (DEN), Gamma Ray log (GR), Compensated Neutron Log (CNL), Photoelectric Factor log (PE), and Resistivity Deep log (RD) logs, while models that incorporated unsupervised clustering used a subset of the GR, DEN, CNL, and U logs. In supervised modeling, we employed several algorithms, such as Light Gradient Boosting Machine (LightGBM), Multilayer Perceptron (MLP), Random Forest (RF), and a hybrid stacking model. The performance results were impressively predictive. LightGBM R² values were, respectively, 0.9600 (train) and 0.9238 (test), indicating considerable predictive power, while MLP obtained R² values of 0.9488 (train) and 0.9309 (test). Random Forest had R² values of 0.9677 (train) and 0.8982 (test), and the stacking model, which overall performed best, achieved R² values of 0.9771 (train) and 0.9342 (test). At the same time, K-means clustering was applied and specifically aimed to analyze the Formation’s heterogeneity by assessing the lithological variation at different depths. The results from both the supervised and unsupervised approaches indicate that there is great potential in the use of machine learning and deep learning techniques to enhance the TOC prediction and advance the comprehension of lithological heterogeneity in the Longmaxi Formation of the Sichuan Basin.