Data-driven lithofacies prediction in complex tight sandstone reservoirs: a supervised workflow integrating clustering and classification models

AbstractLithofacies identification plays a pivotal role in understanding reservoir heterogeneity and optimizing production in tight sandstone reservoirs. In this study, we propose a novel supervised workflow aimed at accurately predicting lithofacies in complex and heterogeneous reservoirs with intercalated facies. The objectives of this study are to utilize advanced clustering techniques for facies identification and to evaluate the performance of various classification models for lithofacies prediction. Our methodology involves a two-information criteria clustering approach, revealing six distinct lithofacies and offering an unbiased alternative to conventional manual methods. Subsequently, Gaussian Process Classification (GPC), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Random Forest (RF) models are employed for lithofacies prediction. Results indicate that GPC outperforms other models in lithofacies identification, with SVM and ANN following suit, while RF exhibits comparatively lower performance. Validated against a testing dataset, the GPC model demonstrates accurate lithofacies prediction, supported by synchronization measures for synthetic log prediction. Furthermore, the integration of predicted lithofacies into acoustic impedance versus velocity ratio cross-plots enables the generation of 2D probability density functions. These functions, in conjunction with depth data, are then utilized to predict synthetic gamma-ray log responses using a neural network approach. The predicted gamma-ray logs exhibit strong agreement with measured data (R2 = 0.978) and closely match average log trends. Additionally, inverted impedance and velocity ratio volumes are employed for lithofacies classification, resulting in a facies prediction volume that correlates well with lithofacies classification at well sites, even in the absence of core data. This study provides a novel methodological framework for reservoir characterization in the petroleum industry.