Abstract: In order to accurately predict the risk level of coal spontaneous combustion，aiming at the limitation of low prediction accuracy of traditional BP neural network model，a prediction model of coal spontaneous combustion risk level coupled with improved grey wolf optimization algorithm（IGWO）and tra‐ ditional BP neural network is proposed.Firstly，a new random dynamic control factor is proposed by using Tent mapping and Levy flight，which improves the development ability and convergence speed of grey wolf optimization algorithm（GWO）.Then，the effectiveness of the IGWO algorithm is verified by six standard test functions.Finally，taking 50 sets of sample data in the goaf as the research object，the IGWO-BP predic‐ tion model of coal spontaneous combustion risk level is established.The absolute value of relative error is used as the evaluation index of model prediction accuracy，and the prediction results of GWO-BP and BP models are compared and analyzed.The results show that the IGWO algorithm has better convergence speed and prediction accuracy than the traditional grey wolf optimization algorithm（GWO）.The prediction effect of IGWO-BP model is better than that of GWO-BP and BP models，and the absolute value of relative error is less than 8%，which greatly improves the prediction accuracy.This research can provide important decision support for the monitoring and early warning of coal mine spontaneous combustion fire.

Key words: coal spontaneous combustion risk grade, IGWO, BP neural network, relative error