The landslide is a certain-uncertain dynamic system. As a explicit behaviors of this complicated nonlinear dynamic system,the deformation of landslide is characterized with many uncertain factors. In order to reduce the limitation of traditional methods, the set pair analysis (SPA) theory was introduced to describe the landslide dynamic system. Based on SPA theory combined with analytic hierarchy process (AHP), a new displacement prediction model is suggested. The existence of limit value of maximum same degree for SPA theory in equal potential condition are also suggested and demonstrated. Based on this demonstration, the new SPA model and probability analysis, some new concepts, such as optimum displacement prediction and potential displacement, are suggested. Furthermore, the potential displacement is also used to set up the quadratic correlation SPA model to describe the dynamic correlation between landslide deformation and reservoir accumulation process. As a test, all these models are used to analyze the newly-happened Liujiatuo landslide in Qingjiang Shuibuya Reservoir. The results indicate following three main conclusions. First, the optimum displacement prediction is reliable and has high precision in short-term prediction ability. Second, the potential displacement is the maximal potential for landslide deformation, which can be regarded as an upper limit of displacement prediction in current status. Furthermore, the variation of it can describe the evolution track for landslide dynamic system in macro level. So, it has guiding significance in early warning for landslide deformation acceleration. Third, the dynamic correlation SPA model is the new way to quantitatively study the relationship between landslide response and reservoir accumulation process. It can be used for quantitative inverse the hysteresis effect on landslide deformation responded to the reservoir accumulation process. The lag time deduced by this back analysis model matches the observed information well. It was proved that these set of models are simple, efficient, and can be promoted in practical engineering.