Rockburst is a main geological disaster influencing underground constructions at the high ground stress area. The prediction of rockburst has become one of the worldwide difficulties in underground works. Formation lithology and the value of ground stress are two basic factors that affect the occurrence of rockburst; and the accurate discrimination of them is directly related to the success of rockburst prediction. According to the measured principal stress data of the engineering site, a numerical calculation model of the region is established by FLAC3D. Using radial basis function neural network, inverse analysis of the initial ground stress field is made. Then based on the rock mechanics in-situ experiments and TPS203 detection technology, formation lithology of the area with high rockburst risks is obtained. Finally, combining inverse analysis data and TSP detection results, the intensity of rockburst within a great distance of the region in front of tunnel face is precisely predicted. The case study shows that the proposed procedure is of high maneuverability; and the rockburst prediction results are well consistent with actual excavation conditions.