Research on Prediction and Early Warning Technology of Gob Spontaneous Combustion Based on RBF Neural Network

Temperature is the most direct indicator to reflect the likelihood of coal spontaneous combustion (CSC). To accurately predict coal temperature, an online observation system for gob temperature and gas was installed in Zhuxianzhuang Mine. Based on this, radial basis function neural network (RBF), particle swarm optimization–radial basis function neural network (PSO-RBF), and error backpropagation neural network (BP) models were established to predict the gob temperature. The reliability of the model predictions was verified using on-site observation data. The results showed that the PSO-RBF model had the best predictive performance, followed by the RBF model, while the BP model had the lowest prediction accuracy. Due to the tendency of the BP model to overfit during training, it exhibited larger errors during the testing phase, leading to reduced predictive performance. Compared to the RBF model, the PSO-RBF model achieved significant improvements in accuracy, with reductions of 17.6% in RMSE, 28.8% in MAE, and 24.2% in MAPE. The results demonstrate that the PSO-RBF model exhibits strong stability and improved universality. Finally, the PSO-RBF model and the online observation system of gob gas were combined to form a new warning system for CSC in the gob. The method achieves accurate prediction and real-time warning of CSC, which has a positive meaning for controlling the CSC in the gob.