Abstract
The construction of a predictive model that accurately reflects the spontaneous combustion temperature of coal in goaf is fundamental to monitoring and early warning systems for thermodynamic disasters, including coal spontaneous combustion and gas explosions. In this paper, on the basis of programming temperature experiment and industrial analysis, 381 data sets of 9 coal types are established, and feature selection was executed through the utilization of the Pearson correlation coefficient, ultimately identifying O(2), CO, CO(2), C(2)H(4), C(3)H(8), C(3)H(8)/CH(4), C(2)H(4)/CH(4), C(2)H(4)/C(3)H(8), CO(2)/CO, and CO/O(2) as input indicators for the prediction model. The chosen indicator data were divided into training and testing sets in a 4:1 ratio, the Particle Swarm Optimization (PSO) methodology was applied to optimize the parameters of the XGBoost regressor, and a universal PSO-XGBoost prediction model is proposed. A tenfold cross-validation method was employed to assess performance of PSO-XGBoost, PSO-RF, PSO-SVR, XGBoost, RF, and SVR models separately, the results underscored the superior predictive accuracy, robustness, fault tolerance, and universality of the PSO-XGBoost model.