Abstract: Landslide displacement is the most intuitive manifestation of landslide deformation, and the prediction of displacement plays a very important role in judging the evolution trend of landslide. Landslide displacement curve is a non-stationary time series affected by various factors. In this paper the trend displacement of Baishuihe landslide in the Three Gorges Reservoir is extracted by the HP filter method. Because of the nonlinear increasing characteristics, the trend displacement which is determined by the characteristics of the landslide is fitted and predicted by polynomial. In the meantime, induced by various factors such as evolution stages, seasonal rainfall, and water level fluctuation, the periodic displacement is trained and predicted by the model of the least squares support vector machine model (LS-SVM). The prediction result of the cumulative displacement is the superposition of the trend term and the periodic term. The results show that the LS-SVM model has high precision in the prediction of monitoring point ZG93 and XD-04, implying that LS-SVM model is of good adaptability in predicting step-like landslide.

Key words: step-like landslide, displacement prediction, time series, filter analysis, least squares support vector machine, trend term, periodic term, Baishuihe Landslide