Abstract: One of the key means of preventing and controlling spontaneous combustion of coal in mines is to predict the natural ignition temperature of coal in the goaf. In order to more accurately predict the natural ignition temperature of the coal in the mining area, coal samples from the 2^# coal seam of Zaoquan Mine were collected for the spontaneous combustion characterization experiment. The data obtained from the experiments were collected to record the changes in the volume fractions of O₂, CO, CH₄, CO₂, C₂H₄ and C₂H₆ gases released from the coal samples during the oxidation heating process. Taking the above marker gas volume fractions as input data and the corresponding oxidation temperatures of the coal samples as output data, a BP neural network-based coal spontaneous combustion prediction model was established using python software by selecting the training set and test set according to the ratio. A novel swarm intelligent dung beetle optimization algorithm was used to optimize the hyperparameters of the BP neural network, and the parameter-optimized DBO-BP prediction model was established, and the prediction results were compared and analyzed with the performance indexes of the particle swarm algorithm (PSO-BP), the genetic algorithm (GA-BP) and the sparrow search algorithm (SSA-BP). The results show that the mean absolute percentage errors (MAPE) in the test phase after hyper-parameter optimization of the BP neural network prediction models by DBO-BP, SSA-BP, GA-BP and PSO-BP are 6.97%, 8.63%, 7.88% and 8.18%, respectively; the regression coefficients R² are 0.976 3, 0.966 8, 0.970 1, 0.969 0, in which the MAPE value of DBO-BP model is the smallest and the R² is the closest to 1. It proves that the DBO-BP prediction model has faster convergence speed, higher solution accuracy, higher prediction accuracy and robustness. Based on the DPO-BP model, the oxidation temperature of the left coal in the goaf of 150202 working face of Zaoquan Coal Mine is predicted. According to the prediction error of the model, the relative error between the predicted temperature and the field measured temperature is 6.97%. The optimized model using the DBO algorithm has a high accuracy in predicting the oxidation temperature of the left coal.