Abstract: There are two main factors that influence the accuracy of landslide displacement prediction. One is the reliability of the prediction model and the other is the quality of field monitoring data. Currently, conventional landslide monitoring technology and evaluation methods have many limitations and shortcomings. In this paper, we propose a new evaluation methodology of the combination of fiber-optic monitoring technology, monitoring data and PSO-SVM prediction model. We use it to predict the deep displacement of Majiagou No. 1 landslide in the Three Gorges Reservoir. Firstly, by analyzing 320 fiber-optics monitoring data of landslide deep displacement, we decompose accumulative displacement into trend and fluctuant components based on the time series method. Then, the trend displacement is predicted with the fitting method. The fluctuant displacement is predicted with the PSO-SVM model. Lastly, the prediction of cumulative displacement is computed with the predicted periodic and fluctuant displacement values. Research results show that the root mean square error is 0.51mm and an average absolute percentage error is 0.37mm, which demonstrate this model has a preferable prediction effect. The predicted total displacement show great consistency with the measured total displacement, with the RSME of 0.54mm and the correlation coefficient of 0.98, respectively. This method can be used to make short-term predictions of landslide deep displacement.