Abstract: For discovering the features of pattern evolution on temporal sequences，based on Takens’ theorem and the definitions of pattern evolution segments，pattern evolution segment sets and frequent pattern evolution，the equivalent relation between the reconstructed pattern evolution and its originality is demonstrated.Methods for finding typical frequent reconstructed pattern evolution and extracting rules from the pattern evolution are proposed.Comparative experiments on periodic data，chaos data and financial data show that the patterns and rules are meaningful and resultful.

Key words: temporal sequence mining, pattern evolution, time-delay embedding

摘要： 为了研究时态序列模式演化特征，在给出模式演化片段、模式演化片段集合和频繁模式演化片段定义之后，基于Takens定理，论证了重构空间内模式演化与原空间模式演化之间的等价性关系；给出了重构后的频繁模式演化范型挖掘方法和频繁模式演化范型生成规则的方法；针对周期、混沌和利率三种不同类型的序列数据进行方法的有效性研究。

关键词: 时态序列挖掘, 模式演化, 空间重构