Abstract: For predicting a mine's slope deformation so as to adopt timely safety measures, this paper proposes a new dynamic synthetical prediction method. With the help of phase-space reconstruction by substituting different lag time, a new data series from the original monitoring data was get for this method. It took the linear optimization method, which on the condition of the minimum sum of squares of prediction residue error, to calculate the weights of different methods' predictions, such as ARMA time series and GM (1, 1) grey theory. The comparison of the results indicates that, the standard deviation of the synthetical prediction is reduced obviously compared with independent approaches; predictions after the Phase-space reconstruction can improve the average precision. And this provides a reliable method for the forecast and precaution for each kind of landslide.