Abstract: In order to realize the efficient prediction and early warning of coal spontaneous combustion, an improved whale optimization algorithm combined with BP neural network (MSWOA-BP) for coal spontaneous combustion temperature prediction model was proposed, and the effectiveness of the model algorithm was verified based on coal spontaneous combustion heating experiments. The predict effects and performance of gray wolf optimization (GWO-BP) model, standard whale optimization (WOA-BP) neural network model and particle swarm optimization (PSO-BP) model were compared and analyzed. It shows that the average percentage error (MAPE) of the four models are 1.735 9%, 2.651 8%, 6.165 5% and 6.570 1% in the training stage, while during the testing phase they are 3.039 3%, 6.072 3%, 6.734 1%, and 7.603 5%. This indicates that the MSWOA-BP prediction model has higher prediction accuracy and stability. MSWOA-BP was applied to predict the temperature at the coal mine site. Comparing the predicted results with the measured temperature, the relative error range is within 2.3% to 12.1%, which realizes the rapid prediction of underground coal mine temperature. It provides a new method for achieving efficient prediction and warning of coal spontaneous combustion.