Abstract: On account of the lack of practicality in landslide deformation prediction models, a dynamic prediction model was established on the basis of time series analysis, system theory and rock-soil rheology theory. The maximum water level, maximum fluctuation, and maximum drawdown speed between two adjacent monitoring processes were chosen as the main external control variables. The trend item of displacement time sequence was extracted and modeled by moving average method and polynomial function, and the nonlinear relation between deviate item and environmental control variables was modeled by support vector machine. The dynamic model can be constructed according to the response of the deformation to the environmental variables. The method is applied to the deformation prediction of landslide in Three Gorges reservoir area. Result shows that this method not only has strong modeling capability but also has high accuracy and can be used in  project programs.

Key words: landslide, deformation prediction, time series analysis, moving average method, support vector machine