Quantitative Evaluation of Deep Coalbed Methane Content: A Logging Data‐Driven Approach

ABSTRACT Accurate evaluation of gas content in deep coal seams is essential for estimating coal‐bed methane reserves and enabling efficient development. Conventional gas content prediction methods primarily target adsorbed gas, rendering them unsuitable for deep coal reservoirs with abundant free gas. This study introduces a novel approach to calculate deep gas content based on logging data. The No. 8 coal seam of the Benxi Formation, situated on the deep eastern margin of the Ordos Basin in the Daning–Jixian Block, serves as the research subject. Multiple regression analysis was employed to model the relationships between logging parameters and Langmuir volume, Langmuir pressure, porosity and water saturation. A logging model for gas content was constructed using the Langmuir equation and the ideal gas law. Four machine learning methods—BP neural network, multiple linear regression, random forest and support vector machine—were utilised to construct a total gas content evaluation model. Errors of the machine learning models were evaluated, and the random forest model was identified as the optimal choice. A comparison between predicted gas content and measured pressure coring results indicates that both the logging and RF models are efficient, accurate and effective. However, the logging model demonstrates superior accuracy compared to the RF model. The logging‐based gas content evaluation method exhibits strong adaptability in the deep sections of the study area, facilitating subsequent geological assessments and resource potential analyses of deep coal reservoirs.