Spontaneous combustion of coal poses a serious threat to mine safety, making accurate temperature prediction essential for early warning and risk prevention. This study develops a robust, data-driven prediction model capable of capturing the nonlinear coupling between coal oxidation gases and temperature evolution. Programmed heating experiments were conducted on coal samples from the Linsheng Mine to analyze the stage-wise evolution of oxidation, pyrolysis, and gas emission characteristics, identifying the critical (60–80 °C) and dry cracking (120–140 °C) temperature ranges. Twelve gas-related indicators—including six single gases (O₂, CO, CH₄, CO₂, C₂H₄, and C₂H₆) and six ratio-based variables (CO/ΔO2, CO/CO2, CO/CH4, C2H6/CH4, C2H4/CH4, and C2H4/C2H6)—were selected as input features. A statistical analysis of their mean, variance, and distribution characteristics was performed to evaluate data variability, followed by a multicollinearity test using the Variance Inflation Factor (VIF) to confirm model stability. To enhance predictive performance, an Improved Tornado Optimizer with Coriolis force (ITOC), incorporating cubic chaotic mapping and a quantum entanglement mechanism, was integrated with a Kernel Extreme Learning Machine (KELM) to construct the ITOC-KELM model. Comparative experiments demonstrate that ITOC-KELM achieves superior prediction accuracy (R2 = 0.9465, MAPE = 21.26%, MAE = 13.97 °C, RMSE = 22.05 °C) and stability compared with seven benchmark models. SHAP (Shapley Additive exPlanations) analysis further interprets the model behavior, revealing that CO, CO₂, and C₂H₄ exert the most positive influence on temperature prediction, while O2 and CH4 show negative or weak contributions consistent with their physical roles in coal oxidation. Validation using independent datasets from the Dongtan and Yuwu coal mines confirms the model’s strong generalization capability and engineering applicability. This work provides an effective framework for intelligent early detection and risk assessment of coal spontaneous combustion. The main challenges lie in acquiring high-quality multi-gas datasets and achieving reliable real-time deployment under complex underground environments. Future work will focus on integrating the proposed model with IoT-based monitoring systems and extending its applicability across diverse geological and mining conditions.

Multi-indicator feature extraction and temperature prediction model for spontaneous coal combustion and oxidation

期刊论文
出版模式：

开放获取
发表日期：

2025年12月4日

查看全文

Xuming Shao,
Xiaojun Zhang,
Wenhao Liu,
Yanchang Li,
Jinzhang Jia,
Zhongxing Li,
Yu Liu

摘要

Spontaneous combustion of coal poses a serious threat to mine safety, making accurate temperature prediction essential for early warning and risk prevention. This study develops a robust, data-driven prediction model capable of capturing the nonlinear coupling between coal oxidation gases and temperature evolution. Programmed heating experiments were conducted on coal samples from the Linsheng Mine to analyze the stage-wise evolution of oxidation, pyrolysis, and gas emission characteristics, identifying the critical (60–80 °C) and dry cracking (120–140 °C) temperature ranges. Twelve gas-related indicators—including six single gases (O₂, CO, CH₄, CO₂, C₂H₄, and C₂H₆) and six ratio-based variables (CO/ΔO₂, CO/CO₂, CO/CH₄, C₂H₆/CH₄, C₂H₄/CH₄, and C₂H₄/C₂H₆)—were selected as input features. A statistical analysis of their mean, variance, and distribution characteristics was performed to evaluate data variability, followed by a multicollinearity test using the Variance Inflation Factor (VIF) to confirm model stability. To enhance predictive performance, an Improved Tornado Optimizer with Coriolis force (ITOC), incorporating cubic chaotic mapping and a quantum entanglement mechanism, was integrated with a Kernel Extreme Learning Machine (KELM) to construct the ITOC-KELM model. Comparative experiments demonstrate that ITOC-KELM achieves superior prediction accuracy (R² = 0.9465, MAPE = 21.26%, MAE = 13.97 °C, RMSE = 22.05 °C) and stability compared with seven benchmark models. SHAP (Shapley Additive exPlanations) analysis further interprets the model behavior, revealing that CO, CO₂, and C₂H₄ exert the most positive influence on temperature prediction, while O₂ and CH₄ show negative or weak contributions consistent with their physical roles in coal oxidation. Validation using independent datasets from the Dongtan and Yuwu coal mines confirms the model’s strong generalization capability and engineering applicability. This work provides an effective framework for intelligent early detection and risk assessment of coal spontaneous combustion. The main challenges lie in acquiring high-quality multi-gas datasets and achieving reliable real-time deployment under complex underground environments. Future work will focus on integrating the proposed model with IoT-based monitoring systems and extending its applicability across diverse geological and mining conditions.