基于可解释性机器学习的煤储层CH4吸附特征研究OA

[Objective]Accurately predicting methane(CH4)adsorption capacity in coal reservoirs is critical to the effi-cient exploration and exploitation of coalbed methane(CBM).Existing studies largely rely on experimental data from a small number of samples and theoretical adsorption models such as Langmuir.However,these studies exhibit time-con-suming and laborious processes,as well as conclusions with limited generalizability,posing challenges to their wide-spread applications across different basins.[Methods]Based on 1150 sets of experimental data on CH4 adsorption in coals,this study investigated five machine learning models,namely artificial neural network(ANN),support vector ma-chine(SVM),K-nearest neighbor(KNN),random forest(RF),and gradient boosting decision tree(GBDT).Then,the SHapley Additive exPlanations(SHAP)algorithm was combined to achieve accurate prediction and interpretation of CH4 adsorption in coal reservoirs.[Results and Conclusion]The stacked ensemble model,combined with the SHAP al-gorithm,reveals that the CH4 adsorption capacity in coals increases with increasing pressure,maximum vitrinite reflect-ance(Rmax),and mass fraction of fixed carbon and decreases with increases in the temperature and the mass fractions of water,ashes,and volatile constituents.Among these factors,Rmax represents the dominant factor controlling CH4 adsorp-tion(contribution ratio:32.7％),while high pressure,low temperature,and high coal rank create the most favorable con-ditions for CH4 adsorption.Based on 2879 sets of data obtained through data expansion using the stacked ensemble mod-el,this study established a multivariate nonlinear formula for calculating the CH4 adsorption capacity in coal reservoirs.This equation was verified using 126 sets of new data from coals of varying ranks in the Ordos and Qinshui basins,yielding a coefficient of determination(R2)of 0.857.The total CH4 adsorption capacity in coal reservoirs across various basins in multiple countries was assessed.The results reveal that the Ordos Basin of China exhibits an average total ca-pacity of 31.87 × 1012 m3 in coal reservoirs at burial depths of less than 5000 m,establishing it as the basin with the greatest resource potential among its various counterparts in countries such as China,the United States,and Australia.Overall,this study enables the accurate prediction of CH4 adsorption capacity in coals,even under limited experimental data or incomplete coal quality data,thereby providing theoretical and technical support for CBM reserve evaluation and the selection of optimal CBM sweet spots.