Abstract: In order to improve the accuracy and timeliness of early warning of coal and gas outburst in working face, a real-time early warning index system based on in-situ stress coefficient, gas pressure coefficient, power coefficient, variation rate and dispersion rate is established by utilizing the change relationship between gas emission characteristics and "three factors" of outburst.The K-means clustering, FOA and RF algorithms are combined to build a real time warning model of coal and gas outburst based on data mining to explore the potential occurrence rules of real time warning indexes and coal and gas outburst, and the optimal warning level is output through intelligent optimization and training of the model.Field application results show that the built early warning indicators better sensitivity, early warning model of the operation time is 0.118 s, in the application of a coal and gas outburst dangerous level 4 hours in advance warning, warning level with the outstanding theoretical results fit well with actual situation, and with the values of K₁, diamond powder quantity S has good consistency, realize real-time and accurate working face of coal and gas outburst early warning.