[Objective] To explore the effect of a compound predicting model in forecasting the deformation time series of landslide in order to provide a new way for the landslide deformation prediction.[Method] Based on support vector machine and BP neural network, a compound predicting model of landslide displacement sequence and rate series was established. The basic information of landslide was analyzed, and extracted. The regression and multi-factor models were constructed by using two kinds of predicting methods, and two time series was predicted. The BP neural network was used to optimize the results.[Result] There was a great correlation between the water level of the landslide reservoir and the two deformation sequence. The stability of the landslide was likely to be weakened by periodic fatigue, and it could be predicted by the compound prediction model of landslide deformation.[Conclusion] The relative prediction error of this study is small, which greatly improves the prediction accuracy and stability of the landslide deformation, and proves the validity of the prediction model.