Research on an Intelligent Early Warning Model for Coal Spontaneous Combustion Based on Stacking Ensemble Learning

Abstract

To address the challenge of accurately predicting the spontaneous combustion temperature of water-immersed and air-dried coal, and to enhance the reliability of early warning systems for coal spontaneous combustion hazards, this paper constructs a coal spontaneous combustion temperature prediction model based on Stacking-based ensemble learning. Based on multi-stage indicator data derived from water-immersed and air-dried coal samples—obtained through programmed-temperature coal spontaneous combustion experiments—Grey Relational Analysis was employed to identify CO, CO₂, C₂H₄, C₂H₆, and the alkane-to-alkene ratio (C₂H₄/C₂H₆) as the core early-warning indicators. Data preprocessing was subsequently completed through normalization and linear interpolation. A two-layer ensemble prediction model was then constructed by selecting Random Forest (RF), Support Vector Machines with Radial Basis Function kernels (RBF-SVM), and Extreme Gradient Boosting (XGBoost) as the base learners for a Stacking framework, with Gradient Boosting Decision Trees serving as the meta-learner; this ensemble model was subsequently validated through comparison against each of the individual models. The results indicate that the RF model exhibits optimal generalization capability, achieving a coefficient of determination (R²) of 0.9027 on the test set. The Stacking ensemble model demonstrates excellent fitting performance on the training set, with a Mean Squared Error (MSE) of 0.3234 and an R² of 0.99; on the test set, it yields an R² of 0.9019 and a Root Mean Squared Error (RMSE) of 23.9473. Its generalization performance is on par with the RF model and superior to the XGBoost model, effectively integrating the strengths of the individual base learners to achieve a dual balance between fitting accuracy and generalization capability. Conversely, the SVM model demonstrates poor adaptability within the specific data context of this study and fails to meet the requirements for coal temperature prediction. The Stacking ensemble learning model constructed in this paper is capable of accurately predicting the spontaneous combustion temperature of water-immersed and air-dried coal, thereby providing reliable technical support for the early warning of spontaneous combustion hazards in the goafs of shallow-buried, closely spaced coal seams.