Gas-Content Prediction in Deep Coal Rock Using a KGA-XGB Model: A Case Study in the Central Ordos Basin, China

Reliable estimation of coal-seam gas content is essential for assessing the resource potential and development prospects of deep coal–rock gas (CRG) reservoirs. However, complex geological conditions limit conventional prediction methods relying only on logging parameters. Taking the No. 8 coal seam of the Benxi Formation in the Ordos Basin as the study target, this study integrates logging and geological parameters and develops a KGA-XGB model for deep coal–rock gas-content prediction. First, K-Means clustering is applied to coal-roof logging responses to construct features representing roof differences and enhance geological information. Then, a genetic algorithm is used to globally optimize key XGBoost hyperparameters. The KGA-XGB model achieves an R2 of 0.84 and an MAE of 2.41 on the test set. Compared with multiple linear regression, R2 increases by 25.37% and MAE decreases by 31.34%. Blind-well validation for wells Y4 and H8 gives mean error rates of 14.39% and 14.63%, respectively, demonstrating good prediction accuracy and generalization. The proposed method provides a practical approach for gas-content prediction in geologically heterogeneous deep coal seams.