Lithology Identification of Deep Coal Seam Based on Machine Learning

Lithology identification is a crucial aspect of coalbed methane exploration and development. Given the intricate composition and lithology of deep coal seams, with frequent interbedding of coal seams, mudstone, and fine sandstone, accurate identification is challenging, particularly due to the similar logging responses of mudstone and fine sandstone. Conventional lithology interpretation algorithms based on crossplots often yield inaccurate results. This study focuses on the deep coal seam in Block SF on the east margin of the Ordos Basin, China. A dataset is compiled based on field logging and coring results. Through correlation analysis, seven key parameters—caliper, compensated neutron porosity, transit-time of compressional wave, gamma ray, deep resistivity log, spontaneous potential, formation bulk density—are selected from the logging data. The comparative analysis results of 16 machine learning algorithms including extra tree, random forest, and xgboost indicate that the extra tree algorithm provides superior fitting and generalization ability, with an F1 score of the classification model reaching 0.798. At the same time, it has high computational efficiency. The predictive outcomes of the model serve as a foundation and reference for determining fracturing intervals and identifying geological sweet spots in subsequent stages. 1. INTRODUCTION Coalbed methane, also known as coal seam gas, is one of the typical clean energy sources. Its extraction holds significant importance for both coal mine safety and strategic energy reserves. China boasts abundant proven coalbed methane resources, with a gas-in-place resource volume reaching 30 trillion cubic meters (Tao et al., 2019). As exploration and development of coalbed methane continue to advance, a trend is emerging from single coal seams to coal series, from low-rank coal to high-rank coal, and from shallow to deep coalbed methane development (Guo et al., 2024). In the eastern SF block of the Ordos Basin in China, coal series formations are widely distributed, exhibiting a rich variety of lithologic combinations and distinct differences in reservoir formation modes (Cai et al., 2022). Particularly noteworthy is the development of thin interbeds of coal and sandstone-shale, posing challenges for reservoir identification and completion design for hydraulic fracturing. Therefore, conducting identification and prediction of lithology in geological formations is of vital significance for deep coalbed fracturing design and favorable area prediction for reservoir formation.