时间序列周期模式挖掘的周期检测方法

摘要

周期是时间序列的重要特征之一,用于精确描述时间序列并预测其发展趋势.在现有周期模式挖掘算法中,周期长度由用户事先定义,忽略了噪声的存在.在ERP度量和时间弯曲算法的基础上,提出一种新的周期长度检测方法.该方法可以在时间轴上实现弯曲,包括延伸和平移.它受噪声干扰的影响较小,实验结果表明其性能优于原有周期检测算法.%Periodicity is an important feature for time series that can be used for describing time series exactly and predicting its development trends. In existing mining algorithms for periodic patterns, the periodicity length is user-specified in andvanc, and the presence of noise is not taken into account. Based on ERP(Edit distance with Real Penalty) measurement and time warping algorithm, this paper proposes a novel algorithm for periodicity length detection, which can realize warp on the time axis including extending and translation. It is less affected by noise interference. Experimental results show that the performance of this algorithm is better than existing periodicity detection algorithms.