基于极值分段特征标识的时间序列分类方法

Time Series Classification Method Based on Feature Identification of Extreme Segmentation

时间弯曲距离受最优路径和距离计算方式限制,累加距离不能有效区分时间序列的类型。标识极值能够获得时间序列的区间性特征,相应的特征标识能够对距离相同但趋势不同的时间序列进行有效分类。提出分析同类时间序列的时间弯曲距离结果获得典型时间序列,根据极值点进行分段,在标准差的基础上形成特征标识。对达到距离要求的目标时间序列和典型时间序列进行标识匹配,最终明确其类型。所提算法解决了时间序列分类过程中时间弯曲距离度量局限性的问题。最后,证明了算法的理论可行性,并给出了其整体流程。实验结果表明,基于极值分段特征标识的时间序列分类方法具有良好的分类性能。

Due to the limitation of optimal path and distance calculation method,cumulative distance of dynamic time warping(DTW) distances cannot classify the types of time series effectively.The interval features of time series can be obtained by identifying extrema.The time series with same distances and different trends can be classified effectively by the corresponding feature identification.In this paper,the typical time series are obtained from DTW results of time series of the same type before segmentation according to extreme points.Then,the feature identification of typical time series is achieved by standard deviation.The target time series that meets the distance requirement and the typical time series are identified and matched,and finally the type of the target time series is determined.The algorithm proposed in this paper solves the limitation of DTW measurement in time series classification.Finally,the theoretical feasibility of the algorithm is proved and the overall flow is given.The experimental results show that the time series classification method based on feature identification of extreme segmentation has good classification performance.