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

Area of Science:

• Mining Engineering
• Chemical Engineering
• Data Science

Background:

• Spontaneous coal combustion is a significant mine safety hazard.
• Accurate temperature prediction is crucial for early warning systems.
• Understanding coal oxidation and gas evolution is key to preventing fires.

Purpose of the Study:

• To develop a robust, data-driven model for predicting coal spontaneous combustion temperatures.
• To capture the nonlinear relationship between coal oxidation gases and temperature.
• To enhance mine safety through intelligent early detection and risk assessment.

Main Methods:

• Programmed heating experiments on coal samples to analyze oxidation and gas emissions.
• Selection of twelve gas-related indicators (single gases and ratios) as input features.
• Development of an Improved Tornado Optimizer with Coriolis force (ITOC)-Kernel Extreme Learning Machine (KELM) model (ITOC-KELM).
• Statistical analysis and multicollinearity testing (VIF) for data variability and model stability.
• SHAP analysis to interpret model behavior and feature influence.

Main Results:

• The ITOC-KELM model achieved high prediction accuracy (R² = 0.9465, MAPE = 21.26%).
• Key gases influencing temperature prediction include CO, CO₂, and C₂H₄.
• The model demonstrated superior performance and stability compared to seven benchmark models.
• Validation on independent datasets confirmed strong generalization and applicability.

Conclusions:

• The ITOC-KELM model provides an effective framework for intelligent early detection of coal spontaneous combustion.
• The study highlights the importance of multi-gas analysis for accurate temperature prediction.
• Challenges include acquiring high-quality datasets and real-time deployment in underground environments.