Prediction of Gas and Water Production of Coalbed Methane Wells by Integrating Light Gradient Boosting Machine and Multioutput Regressor

Summary The production of coalbed methane (CBM) significantly supplements unconventional natural gas supply. Accurate prediction of CBM well production is crucial for optimizing development strategies and evaluating project economics. Existing studies typically rely on historical production data and some influencing factors to forecast future CBM production. However, few studies directly predict the entire production dynamics solely based on influencing factors. To address this gap, we propose an improved light gradient boosting machine (LightGBM) model that integrates the multioutput regressor (MOR), capable of predicting the daily gas and water production of CBM reservoirs by leveraging static reservoir geological parameters and dynamic bottomhole pressure (BHP) data. Experimental results demonstrate that the proposed model accurately predicts CBM well production. For daily gas prediction, the determination coefficients (R2), mean squared errors (MSE), and mean absolute errors (MAE) are 0.952, 0.77 × 10−4, and 3.60 × 10−3, respectively. For daily water prediction, the R2, MSE, and MAE are 0.935, 0.84 × 10−4, and 3.34 × 10−3, respectively. Compared with conventional machine learning models, our proposed method exhibits superior prediction accuracy and robustness. It not only eliminates dependence on historical production data but also significantly reduces computational costs and enhances prediction efficiency compared with numerical simulation methods. Moreover, our approach holds potential for extension to other unconventional resources like shale and tight oil/gas reservoirs by adapting corresponding training data.