基于模态重构与多维评价的时间序列趋势提取

Trend feature extraction method for time series basedon mode reconstruction and multidimensional evaluation

为了准确提取时间序列的趋势特征,提出一种基于模态重构与多维评价的时间序列趋势提取算法。定义重要点作为时间序列分段点的候选集,运用自适应噪声的完备经验模态分解方法对时间序列进行分解和模态重构得到全局因子,使用全局因子度量重要点在整体维度上的重要程度,给出特征因子和边界因子的定义并分别用来度量重要点在单点维度和局部维度上的重要程度,根据3个评价因子综合评价重要点来选取分段点。仿真实验结果表明,该方法具有良好的去噪能力,在相同压缩率情况下的拟合精度比现有方法高,在对心电图趋势提取的实验中也验证了方法的有效性。

Time series are characterized by“high dimension and mass”.Due to the existence of noise interference,it is difficult to analyze the overall trends of time series.To extract the trend characteristics of time series accurately,this paper proposes a time series trend extraction algorithm based on mode reconstruction and multidimensional evaluation.First,the important points are defined as the candidate set of the segment points.Then,CEEMDAN(complete ensemble empirical mode decomposition with adaptive noise)is used for mode decomposition and mode reconstruction of time series to obtain global factors,and the global factors can evaluate the importance of important points in the dimension of overall time series.Next,characteristic factors and boundary factors are defined and used to evaluate the importance of important points in the dimension of single-point and the dimension of local time series,respectively.Finally,the three evaluation factors are used to comprehensively evaluate important points to select segment points.Simulation experiment results show that the proposed method has good denoising ability,and its fitting accuracy is higher than that of existing methods under the same compression ratio.The effectiveness of the proposed method is also verified in the experiment of trend extraction of ECG.