Abstract: To solve the problem of losing sequence fluctuation information when analyzing time series data with traditional symbol aggregation approximation methods,a new Symbol Aggregation approximation method of time series combined with Volatility (SAX_VOLA) was proposed.In this method,the volatility index to quantify the fluctuation range and trend information was introduced,and the symbol vector fused with volatility was used to approximate the subsequence.On this Basis,a new distance metric was proposed to measure the similarity of time series and to perform classification learning.The proposed method was tested on several open data sets,and the experimental results showed that the better classification accuracy than traditional symbolic aggregation approximation methods was achieved,especially when times series had obvious local fluctuations or obvious upward and downward trends.

Key words: time series analysis, symbol aggregation approximation, production data analysis, classification learning

摘要： 为解决传统符号聚合近似方法分析时序数据时丢失序列波动形态信息的问题,提出一种融合波动信息的时间序列符号聚合近似方法。该方法在传统符号化方法的基础上定义波动率指标来同时量化时间序列的波动幅度和变化趋势信息,用融合波动率的符号矢量近似刻画子序列,在此基础上给出一种新的时间序列距离度量方法。以此度量方法为基础,提出时间序列的相似性计算和分类方法,并在公开数据集上进行了分类学习实验。实验结果表明,所提方法在绝大部分数据集上获得了较传统符号聚合近似方法更好的分类准确率,尤其在时间序列具有明显的局部波动或明显的上升、下降趋势时。

关键词: 时间序列分析, 符号聚合近似, 生产线数据分析, 分类学习