时间序列本身具有高维、高噪声的特点.在进行相似性度量之前,需要对序列进行特征表示.针对时间序列相似性度量工作中,使用分段线性表示方法对序列进行特征表示,分段拟合效果依赖于划分粒度,若分段数和分段点选取不当,可能导致拟合效果不佳、难以准确反映序列整体形态趋势的问题,提出一种新的基于趋势的相似性度量方法.该方法将经验模态分解方法与分段线性表示方法相结合,首先用经验模态分解方法过滤细节信息,提取序列的主要形态趋势,得到趋势拟合序列.在此基础上,再用分段线性表示方法对趋势拟合序列进行分段表示,减少拟合结果对划分粒度的依赖性.最后给出序列的分段向量距离计算方法,对趋势分段序列计算加权向量距离,得到不同序列之间的相似性.仿真实验表明,该算法稳定有效、对噪声不敏感.%The time series itself has characteristics of high dimension and high noise. It is necessary to represent the sequence before the similarity measure. When using piecewise linear representation method for feature representation,piecewise fitting results depend on the partition granularity. If the segmentation number and segmentation points are not proper selection,which may lead to poor fitting and could not accurately reflect the overall trend of the sequence form. Therefore,aiming at the problem,a new method of similarity measure-ment based on the trend is proposed which combines the empirical mode decomposition with piecewise linear representation. Firstly,filte-ring details by empirical mode decomposition,extracting main morphological trend of the sequence,the trend fitting sequence is gained. On this basis,use of piecewise linear representation for fitting the trend sequence,the dependence of the fitting result on the partition gran-ularity is reduced. Finally,the calculation method of piecewise vector distance is given. The similarity between different sequences can be obtained by calculating the weighted vector distance of the trend segment sequence. Simulation results show that the proposed algorithm is stable and effective,and not sensitive to noise.
展开▼
