时间序列特征提取方法研究综述

Overview of Feature Extraction Algorithms for Time Series

随着物联网、大数据和人工智能等技术研究和应用的蓬勃发展,各类时间序列数据不断涌现.时间序列数据特征是表象,内在蕴含着丰富的领域知识,如何高效分析时间序列特征模式,提取可辨识的时间序列特征,挖掘数据蕴含的规律,正成为业界研究的热点.本文首先介绍时间序列概念,综述了时间序列分类、聚类和预测三方面研究的最新进展;然后从时间序列特征提取方法的形状特征、时间依赖特性、序列变换特征3方面,详细分析和比较机器学习方法在时间序列问题上的研究情况,最后基于当前时间序列特征提取方法的发展趋势,对时间序列特征提取方法的未来发展做出展望.

With the development of machine learning,new technologies continue to emerge in the field of time-series.How to efficiently analyze the internal patterns of time series and extract the identifiable characteristics of time series is becoming a research hotspot.This article first introduces the latest developments in TS research,then analyze and compare the research situation of machine learning methods on time series in detail from the aspects of waveform extraction,time-dependent characteristics,and sequence transformation of the feature extraction algorithm,and finally based on the development trend of the current time series feature extraction algorithm,the future development of the time series feature extraction algorithm is prospected.