A semi-supervised coarse-to-fine approach with bayesian optimization for lithology identification

Abstract Lithology identification is critical in the interpretation of well-logging data for petroleum exploration and development. However, limited availability labeled machine learning model training can lead to compromised accuracy lithology classification models. Here, we propose a semi-supervised overcome this challenge. Our framework consists Bayesian optimization tuning ensemble algorithms, including random forest, gradient boosting decision tree, extremely randomized trees, adaptive boosting, establish high-quality baseline learning. We also employ self-training strategy increase number samples set use predicted label with highest confidence as pseudo-label reduce accumulation deviation caused by incorrect pseudo-labels. coarse-to-fine improves rock accuracy, particularly sandstone. Testing our on from two real regions, found that ExtraRF-based HGF area performs best, maximum 91.6 $$\%$$ % , which 5 higher than original without using pseudo-labeling techniques.