Study on a Data-Driven Evaluation Model for Methane Adsorption Capacity in Coal Reservoir

Accurate evaluation of methane adsorption capacity in coal reservoir is a crucial aspect of coalbed methane (CBM) geological resource assessment. Traditional laboratory evaluation methods are costly and time-consuming, while data-driven approaches such as neural networks often suffer from model complexity, lack of interpretability, and difficulties in practical application. This study proposes a data-driven evaluation model for methane adsorption capacity (VL) in coal. The model is based on the Alternating Conditional Expectation (ACE) method and is developed using 308 data samples obtained from literature statistics and experimental measurements to construct and validate the ACE​evaluation model. The sample data cover a moisture content range of 0%–7.47%, an ash content range of 1.2%–22.31%, a fixed carbon content range of 11.98%–93.04%, a maturity range of 0.5–4.56, a temperature range of 273.5 K–349.5 K, and a vitrinite content range of 0.2%–98.3%. The methane adsorption capacity of coal rock samples follows the order: ash-free samples > dry samples > equilibrium moisture samples. A total of 299 randomly selected samples were used as the training set, and the resulting ACE model was expressed as an explicit sixth-order polynomial. The remaining nine samples were used for validation, and the model demonstrated a correlation coefficient greater than 0.99, a mean absolute error (MAE) of 0.16, a mean absolute relative error (MARE) of 1.09%, and a variance of 0.03.This study provides theoretical support for the rapid and accurate field prediction of methane adsorption capacity in coal rock, promotes the digitalization of rock physical properties, and offers significant technical guidance for efficient CBM resource assessment.