Comparative study on prediction of coal seam gas extraction based on Extreme Gradient Boosting and random forest model improved by optimization algorithm

Gas, a silent and deadly hazard in coal mines, poses a significant risk of coal seam gas outbursts and excessive emissions. Effective coal seam gas drainage is crucial for mitigating these risks. This study focuses on the coal seam characteristics of the 21 601 transports gallery in the Qinglong coal mine, selecting drainage stage, negative pressure, and concentration as input variables, with the volume of gas drainage as the output variable. We have integrated the XGBoost (Extreme Gradient Boosting) and random forest (RF) algorithms with Bayesian, Sparrow, Scarab, and Gorilla optimization algorithms-establishing a composite model for predicting coal seam gas drainage volume. Our research indicates that the predictive performance of models optimized by these algorithms surpasses that of other models. Specifically, the XGBoost algorithm outperforms the RF algorithm in predicting coal seam gas drainage volume. Among the optimization algorithms tested, the OP (Bayesian optimization) algorithm demonstrated the poorest fit and highest error rates. In terms of validation set performance, the XG-GTO (Gorilla and XGBoost combined algorithm) composite model excelled, with metrics of MAE (mean absolute error) = 0.217 82, MAPE (mean absolute percentage error) = 0.1149, MSE (mean square error) = 0.082 153, RMSE (root mean square error) = 0.286 62, and R-2 (coefficient of determination) = 0.920 59. Furthermore, the Shapley additive explanations revealed that drainage concentration has the most significant impact on gas drainage. This study not only furnishes robust data support for the construction of coal mine big data but also holds substantial value for the development of intelligent coal mine systems and the enhancement of intelligent gas drainage technologies.