Abstract: Rock burst is one of the main geological hazards in underground constructions at the high stress areas, so it is very important to be able to predict rock burst in an accurately and timely manner. Rock burst prediction has developed from using a single indicator to using multiple and comprehensive indicators. Radial basis function (RBF) neural network is an integrated prediction method. In RBF neural network design, network is designed as high dimensional space curve fitting, in which lowdimensional model inputs is transformed to highdimensional space through hidden layer.In this way,the nonlinear mapping between the input vectors and output vectors is established. Based on several theoretical criteria specific to the project,the established RBF neural network model is applied to rock burst prediction for a real project.The results of prediction are consistent with the actual situation.

Key words: Rock Burst, RBF Neural Network, Underground Caverns