Prediction of coal temperature during spontaneous combustion based on the fusion of primary and auxiliary multi-gas indicators and GA-GWO-SVR: a case study of coal mines in Ningxia, China

Accurate prediction of coal temperature is important for the early warning, prevention, and control of goaf fires. To address information redundancy among multi-gas indicators and the limitations of single-algorithm optimization of SVR hyperparameters, a coal-temperature prediction method combining primary–auxiliary indicator fusion with GA-GWO-SVR was developed. Coal samples from Ningxia, China, were tested using a two-tonne large-scale coal oxidation system. Analysis of gas evolution and its correlation with coal temperature identified CO, C 2 H 4 , and C 2 H 4 /C 2 H 6 as primary indicators, and O 2 , CO 2 , CH 4 , C 2 H 6 , and derived ratios as auxiliary indicators. These indicators provided the basis for a five-level early-warning system. For model construction, all 10 input features were first Z-score normalized; PCA was then applied only to the seven auxiliary indicators, whereas the three standardized primary indicators were retained directly, preserving direct information from the primary indicators while reducing redundancy and multicollinearity. A sequential GA-GWO strategy coordinated GA-based global search and GWO-based local refinement to optimize C, γ, and ε. The model achieved an MAE of 1.66 °C, an RMSE of 2.55 °C, a MAPE of 2.85 %, and an R 2 of 0.9944 on the test set, showing the best overall performance among the evaluated models. Validation using the Wangwa coal sample excluded from model development yielded an RMSE of 3.10 °C and an R 2 of 0.985, supporting applicability to an unseen coal sample. Field application at Wangwa No. 2 Coal Mine showed that the proposed method could convert goaf gas-monitoring data into predicted coal temperatures and early-warning levels, thereby providing guidance for graded responses involving goaf fire-prevention and control measures and supporting safe mine production.