CAJ | 학술논문

时间序列的表示是时序数据挖掘的一个重要问题.重要点的分段表示法(IP)是目前应用最为广泛的时间序列特征提取方法之一,具有较好的数据压缩和去除噪声能力,但参数的选择对时间序列的近似效果有很大的影响而且难以找到重要的转折点.基于多分辨率的重要点检索分段方法(MIP)也是一种时间序列特征提取方法,该方法能很好地近似时间序列,但检索次数难以确定且运行效率比较低.为了改进以上两种方法的缺陷,提出了一种新的基于重要点的多分辨率检索表示法(MRIP).实验结果表明,与基于重要点分段方法相比,该方法误差更小,具有很好的压缩率,并能去除噪音干扰;与基于多分辨率的重要点检索分段方法相比,能较好地确定检索次数的范围,在近似效果相当的情况下,运算效率更高.
시간서렬적표시시시서수거알굴적일개중요문제.중요점적분단표시법(IP)시목전응용최위엄범적시간서렬특정제취방법지일,구유교호적수거압축화거제조성능력,단삼수적선택대시간서렬적근사효과유흔대적영향이차난이조도중요적전절점.기우다분변솔적중요점검색분단방법(MIP)야시일충시간서렬특정제취방법,해방법능흔호지근사시간서렬,단검색차수난이학정차운행효솔비교저.위료개진이상량충방법적결함,제출료일충신적기우중요점적다분변솔검색표시법(MRIP).실험결과표명,여기우중요점분단방법상비,해방법오차경소,구유흔호적압축솔,병능거제조음간우;여기우다분변솔적중요점검색분단방법상비,능교호지학정검색차수적범위,재근사효과상당적정황하,운산효솔경고.
Time series data representation is one of the important problems of time series data mining.Piecewise linear representation for time series based on important point(IP)is one of the most widely employed methods of feature extraction for time series.This method can compress time series much and remove noises in time series.However,the selection of the parameter has great effect on the result of approximation for time series.Also,the method is hard to find the turning points.Multiresolution important point retrieval method(MIP)for time series is another method of feature extraction for time series.This method performs well in the result of approximation for time series.But choosing the number of retrieval is difficult and the speed is also lOW in this method.In order to rectify the shortcomings of the above two methods.a novel multi-resolution retrieval method based on important point(MRIP)for time series representation is proposed in this paper.Compared with IP,the new method can approximate to raw time series more precisely,compress time series more validly,and remove noises more effectively.Compared with MIP,the new method has smaller range in number of retrieval,is higher in speed and has almost no degrade in approximation for time series.Experimental results proved what mentioned above.