【摘要】 中心时间序列表明了一个时间序列集合中的公共特征,是时间序列聚类的重要手段。提出了一个利用动态规划求解两条时间序列DTW中心的方法,即以最小化中心序列到两条样本序列的DTW距离平方和为目标,递归求解最优解。在此基础上,给出了基于中心与样本匹配度的剪枝方法,降低了时间复杂度。并在理论上证明了该方法可以获得最优解。实验结果显示,相比于DBA算法,该算法能够获得更小的DTW距离平方和,并且有更好的鲁棒性。更多还原

【Abstract】 The central time series plays an important role in time series clustering,which indicates the common features of time series.We proposed dynamic planning approach called DPSSD to calculate central time series of two time series.The approach is recursive based on the minimizing sum of squares of DTW(SSD)distance from central series to two sample series.Degree-pruning was also introduced to decrease the algorithm time complexity.The proposed algorithm was proved theoretically.It can achieve the optimal solution.In the experiments,the results illustrate that our approaches have much better performance and robustness than DBA,which is measured by SSD.更多还原