Abstract: The ground surface subsidence of underground cave is serious geological hazard. In order to predict ground surface subsidence of underground cave a nonlinear subsidence model of underground cave is established according to stochastic medium method. An accelerated hybrid generic algorithm is put forward based on the DFP method and modified float genetic algorithm, and applied to optimizing parameters of nonlinear subsidence model. Results show that the accelerated hybrid generic algorithm is practical, efficient and superior to other methods.