Deep learning framework based on ITOC optimization for coal spontaneous combustion temperature prediction: a coupled CNN-BiGRU-CBAM model.

Coal spontaneous combustion (CSC) poses a significant safety hazard in coal mines, requiring effective prevention and control strategies. Accurate temperature prediction, crucial for assessing coal oxidation stages and combustion risk, underpins early warning systems. This study analyzes programmed heating experimental data from Dongtan Mine coal samples and integrates the coal oxidation-pyrolysis coupled reaction mechanism. Pearson correlation analysis identified six key gas indicators-O₂, CO, C₂H₄, CO/ΔO₂, C₂H₄/C₂H₆, and C₂H₆-highly correlated with spontaneous combustion temperature. Based on these variables, a deep learning framework combining an Improved Tornado Optimization with Coriolis force (ITOC) strategy and a CNN-BiGRU-CBAM model is proposed. The ITOC algorithm incorporates cubic chaotic mapping initialization, quantum entanglement, and Coriolis force perturbation to enhance global optimization. Comparative experiments with five heuristic algorithms demonstrate ITOC's superior accuracy and convergence stability. Key CNN-BiGRU-CBAM hyperparameters-learning rate, BiGRU neuron count, and convolutional kernel size-were jointly optimized by ITOC, resulting in optimal values of 0.0093, 108 neurons, and 8.54, respectively. The dataset was split into training, validation, and test sets at an 8:2:1 ratio. Performance evaluation against benchmark models shows the proposed framework achieves a test set R of 0.9738, MAPE of 4.1254%, MAE of 6.2740, and RMSE of 12.4735. Validation on coal faces in Shandong, Shanxi, and Shaanxi mines confirmed strong generalization and engineering adaptability, with predicted temperature ranges closely matching measurements. The ITOC-CNN-BiGRU-CBAM model offers a promising theoretical and practical approach for intelligent early warning and precise prevention of CSC hazards.