Abstract: The phase space reconstruction technique and the computation method of chaotic attractor’s correlation dimensions are introduced. The forecasting principle and the forecasting model based on the peculiarities of chaotic attractors are expounded, deficiencies existing in the present model are disclosed, and modification opinions and a concrete algorithm are put forward. Then, using the time series from Lorenz system as samples for forecasting, the results of three forecasting models are presented for comparison, and some problems remained to be resolved are pointed out.