Application of nonlinear time series analysis in slope deformation analysis and forecast

Abstract

The slope is a nonlinear dissipative dynamic system, which is controlled by the condition of rock mass and influenced by the terrain, groundwater, earthquake and human projects. So its deformation takes on nonlinear feature. In this paper the method of phase space reconstruction is discussed including the method of mean mutual-information used to determine the delay time-delay and the method of the nearest neighbors to the embedding dimension. Based on the nonlinear feature, the radial basis function is selected to build neural network for forecasting the deformation and compared with the BP neural network. The results show that the radial basis function model has well generalization ability. It is much better than BP network in the convergence speed and predicting accuracy. © 2008 Science Press Beijing and Springer-Verlag GmbH Berlin Heidelberg Geotechnical Engineering for Disaster Mitigation and Rehabilitation.