Prediction of coal structures and its gas-bearing properties based on geophysical logging parameters: A case study in Anze block, China.

This article focuses on the identification and characterization of coal structures in the Anze block of the southern Qinshui Basin, China, and their implications for coalbed methane (CBM) development. Using geophysical logging data and machine learning methods—including random forest algorithm (RFA), radial basis function neural network (RBF), and long short-term memory neural network (LSTM)—the study achieved high accuracy in predicting four coal structure types: intact, cataclastic, granulated, and mylonitic coals. Results indicate that cataclastic coal, controlled mainly by tectonic activities and burial depth, is the most prevalent and favorable for CBM extraction due to its higher gas content and better hydraulic fracturing performance compared to granulated coal. The study also confirms that the gases are primarily thermogenic, with an average δ¹³C(CH₄) value of −37.51‰, and highlights the importance of integrating multi-parameter logging data and machine learning for efficient coal structure identification in CBM reservoirs.