在以往混沌时间序列的分析中，往往只关注其动态物理机制的变化而忽略了其本身所含的信息量的刻画。事实上，相空间重构后原来时间序列所含的信息量也会随着改变，同时考虑到秩次集对分析是利用历史集合确定当前集合的后续值，所以需要保证历史集合的信息量尽可能大。因此本文加入了对重构相空间中信息量的研究，利用新定义的最大信息量下的最小预测误差函数选取嵌入维，建立了IC-SPA（Information-max of chaos SPA analysis）模型，并用IC-SPA模型预测了太原，密云，石家庄每月的极端最高气温，并和传统的预测误差最小法的SPA模型，BP模型，AR模型进行比较，IC-SPA模型都达到了很好的效果。IC-SPA模型的预测相比较其他模型，其MRE太原减小了40.45%，43.02%，58.68%, 密云减小了37.89%，55.29%，56.04%, 石家庄减小0%, 75%，52.11% ，说明IC-SPA模型有很好的预测能力。

For the traditional chaos time series analysis, it usually focuses on the dynamic change of physical mechanism, but ignores the description of information which contains its own characterization. In fact, with the phase space reconstruction, the information will also change. With considering the prediction theory of Set Pair Analysis (SPA)—namely using the historical sets to predict the next value of current set, the history sets should contains great information. Based on that, I add the information factor to phase space reconstruction, provide a new way to calculate the embedding dimension，and the information chaos SPA analysis (IC-SPA) model is established. In the new model, the information in reconstructed phase space is max, which provide good history sets to prediction. Three cases in forecasting extreme temperature in Taiyuan, Miyun, Shijiazhuang are taken to examine the performance of IC-SPA model. The results indicate that the mean relative error (MRE) decreased 40.45%，43.02%，58.68% in Taiyuan, 37.89%，55.29%，56.04% in Miyun, and 0%, 75%，52.11% in Shijiazhuang, respectively, compared with rank set pair analysis(R-SPA), Back-Propagation (BP) neural network model and autoregression (AR) model.