Objective  The mining of deep coal seams in the Ximeng mining area within the Ordos Basin is subjected to complex environmental conditions like high in-situ stress, large water inflow, and severe air leakage, which lead to the encountered with elevated risks and difficult prediction of coal spontaneous combustion and posing challenges in predicting spontaneous combustion.

Methods Coal samples from the Yingpanhao and Shilawusu coal mines in the Ximeng mining area were selected for temperature-programmed spontaneous combustion experiments to determine the characteristic parameters of coal spontaneous combustion under different moisture contents and sulfur mass fractions. Based on these parameters, as well as with coal quality parameters from proximate analysis, a prediction database was established. Then, the hyperparameters of the random forest (RF) model were optimized using the crested porcupine optimizer (CPO) algorithm. Accordingly, the CPO-RF model was constructed to predict the degree of coal spontaneous combustion.

Results and Conclusions  The results indicate that the coal samples from the Yingpanhao and Shilawusu coal mines showed similar laws of variations in gas concentrations and oxygen consumption rates during oxidative heating. CO was identified as the dominant indicator gas, appearing initially at a temperature of about 30℃. The amount of gas produced increased with the sulfur mass fraction. However, as the moisture mass fraction increased, it decreased initially and then increased. The coal spontaneous combustion manifested critical temperatures ranging from 67.5℃ to 70.5℃ and dry cracking temperatures from 113.5℃ to 115.4℃. The optimal tree depth and tree count of the RF model were automatically identified using the efficient global search capability of the CPO algorithm, avoiding local optimal solutions caused by improper settings and thus enhancing the generalization and robustness of the model. The constructed CPO-RF model significantly improved the prediction accuracy of coal spontaneous combustion. As a result, the predicted temperatures based on the test set coincided well with the actual values, with a mean absolute error of 0.762℃, a root mean square deviation of 1.014, and a coefficient of determination of 0.999 4. The comparison between the predicted results of the CPO-RF model and the characteristic temperatures of coal spontaneous combustion enabled the efficient discrimination of the risks of coal spontaneous combustion. Based on this, targeted fire prevention and extinguishing methods can be adopted. The results of this study serve as a reference for preventing coal spontaneous combustion for deep coal mining in mining areas.