一种基于符号化的时间序列预测方法

Time Series Prediction Method Based-on Symbol Sequence

本文提出一种基于符号化方法对时间序列进行预测。该方法利用矢量拟合来表达时间序列走势的形态，采用聚类算法对形态进行聚类，根据聚类结果得到符号序列，并用不完全抽取方法来抽取序列模式。预测时，根据学习得到的模式集对新序列做出预测分析。对导航位置误差数据实验表明，该方法可以对时间序列进行较好预测。

A new method for time series based on symbolic sequences is presented. The method uses vector fitting to represent the forms of the trends of time series and applies cluster method to classify these forms. Then the symbolic sequences are obtained and the patterns of sequences are extracted by using half-baked patterns extracting algorithm. Experiments show that it gives good results to forecast navigation error time series.