Lithology Identification and Prediction of Coal-bearing Strata While Drilling Using Deep Learning: A Review

Lithology identification and prediction while drilling are essential for safe and efficient drilling in coal-bearing strata, where frequent coal-rock transitions and complex geological structures impose high demands on real-time strata sensing. Recent advances in machine learning and deep learning have provided new opportunities for automatic lithology identification and prediction using geophysical, drilling, image, and multi-source data. This paper reviews recent progress in lithology identification and prediction while drilling for coal-bearing strata. The review first summarizes the characteristics of conventional rotary drilling and directional drilling, together with the available data sources and preprocessing requirements. It then discusses representative machine learning and deep learning methods for lithology identification based on geophysical data, drilling data, and multi-source data. For lithology prediction, particular attention is given to data imbalance, label scarcity, few-shot learning, and multi-source fusion. Finally, the major challenges and future directions are analyzed, including geological data heterogeneity, minority-class recognition, model generalization, interpretability, and real-time field deployment.