Abstract: The in-situ monitoring data of the surrounding rock displacements reflect the changing of mechanical situation of surrounding rock-supporting system. To overcome the excessive learning problem of ANN, a new method of PSO LS_SVM to forecast the nonlinear displacements of surrounding rock in underground engineering is presented based on monitoring data. Particle swarm optimization is used to choose the parameters of support vector machine, which can avoid the man-made blindness and enhance the efficiency and capability of forecasting. The method can rolling forecast the surrounding rock deformations based on monitoring data, in order to adjust the supporting schemes dynamically and ensure the stability of the cavern. It is used to forecast the surrounding rock convergent deformations of Qingjiang Shuibuya Underground Powerhouse; and it is shown that the method is feasible and precise.

Key words: surrounding rock deformation, time series prediction, particle swarm optimization, support vector machine