Spontaneous coal combustion temperature prediction based on an improved grey wolf optimizer-gated recurrent unit model

Abstract

Predicting the spontaneous combustion temperature of coal is crucial for early warning systems, as it is significantly influenced by changes in temperature and gas concentration. Based on programmed temperature rise experimental data from the Dongtan coal mine, Spearman’s rank correlation coefficient analysis and coal oxidation pyrolysis composite reaction identify six gas indicators closely related to coal temperature. An improved grey wolf optimizer (IGWO) is proposed and combined with a gated recurrent unit (GRU) model to predict spontaneous coal combustion temperature. Model improvements include enhancing global exploration capability using chaotic mapping, improving local search accuracy with a nonlinear convergence factor, and introducing an optimal memory retention strategy to optimize convergence speed and stability. The IGWO's capability and adaptability are verified by comparison with eight common heuristic optimization algorithms. The GRU model adopts a double-layered structure, with each layer containing 20 GRUs. To further improve prediction accuracy, three hyperparameters—number of hidden units, initial learning rate, and maximum number of training epochs—are optimized (85, 0.165451, and 66, respectively). The dataset is randomly split into a training and test set (7:3). The IGWO-GRU model is compared with 13 reference models and achieves an R² of 0.9510, root mean square error of 16.0503, and maximum relative error of 0.06 on the test samples, significantly outperforming other models. Finally, the IGWO-GRU prediction model applies to a programmed temperature rise experiment for spontaneous coal combustion, demonstrating superior prediction accuracy and practical application potential.