In the process of establishing fuzzy time series model, the data pre-process and the optimization of the fuzzy rules are the two key factors which influence the precision of predication. Therefore this paper proposes a stabilization algorithm based on principal component analysis(PCA). Firstly, time series are changed into stationary time series by augmented dickey-fuller tests. Then, the fuzzy sets are set up, the generalized covariance matrix of fuzzy relationships is built, and its eigenvalues and eigenvectors are calculated. Furthmore, fuzzy rules are optimized according to the cumulative of the eigenvalues, and forecasting results for real time series are obtained. Finally, the experiment results show the feasibility of the proposed algorithm.