基于乘积量化编码器的时间序列数据降维

Time Series Data Dimension Reduction Based on Product Quantization Encoders

时间序列数据维数高,导致在原始时间序列上执行数据挖掘任务效率很低。因此,在对时间序列执行不同的数据挖掘任务之前,需要对时间序列降维。本文基于乘积量化(ProductQuantization)编码器设计编码以及解码,通过选择时间序列中的重要点来将时间序列投影到低维空间,该降维方法能对时序数据进行自适应,并能保留数据集上的相关信息,从而对信息进行泛化。最后,为验证本文所提数据降维方法的有效,在两种时间序列数据集上进行实验。实验结果表明:本文方法在对数据进行高度压缩的同时保留更多特征信息。

The high dimension of time series data leads to the low efficiency of data mining on the original time series. Therefore,before performing different data mining tasks on time series, it is necessary to reduce the dimension of time series. This paper is based on product quantization encoder designs encoding and decoding, and projects the time series to the low dimensional space by selecting the important points in the time series. The dimensionality reduction method can adapt to the time series data and retain the relevant information on the data set, so as to generalize the information. Finally, in order to verify the effectiveness of the data dimensionality reduction method proposed in this paper, in two time series Experiments were performed on data sets. The experimental results show that this method can highly compress the data and retain more feature information.