基于多尺度模糊熵的时间序列特征提取算法

Feature extraction algorithm of time series based on multi-scale fuzzy entropy

针对现有时间序列数据维度高、时间序列学习算法时间复杂度高以及传统的时间序列数据特征表示方法不能提取时间序列的重要数据特征等问题,文章提出一种基于多尺度模糊熵的时间序列特征提取算法。该算法依据不同的数据集对各段时间序列的均值等特征进行提取,同时对模糊熵进行粗粒化,并利用多尺度模糊熵提取时间序列的分段特征。在真实数据集及UCR时间序列数据库上的实验结果表明,该特征提取方法表现出良好的分类能力,具有有效性和可行性。

Due to the high dimensionality of existing time series data,time series learning algorithms have the problem of high time complexity,and traditional feature representation methods for time series data cannot extract important data features of the time series.Therefore,the article proposes a time series feature extraction algorithm based on multi-scale fuzzy entropy.Extract the mean and other features of each time series based on different datasets,coarsen the fuzzy entropy,and use multi-scale fuzzy entropy to extract segmented features of the time series.The experimental results on real datasets and UCR time series databases show that this method exhibits good classification ability,verifying the effectiveness and feasibility of this feature extraction method.