Research On Similarity Measurement Method Of Time Series

With the continuous advancement of science and technology,t he era of big data is gradually coming.As an aspect of big data,time series research has attracted more and more attention.The essence of time series is to arrange the values of the same attribute in a set of data points in chronological order.With the increase of data,how to dig out similar related information from a large amount of time series information and contribute to human development has become particularly important.Since the time series has the characteristics of large dimensions and a lot of noise,it is difficult to measure the similarity directly on it.Therefore,we usually use a variety of methods to deal with time series.Before using feature representation method to measure the similarity of high-dimensional time series,dimension red uction is carried out.The distance between time series is calculated by clustering method.Although there are some effective methods,there is still room for improvement.This paper analyzes the literature achievements of time series in the field of data mining in recent years,and makes a more in-depth study on the similarity measurement of time series.Firstly,an adaptive weight DTW method based on segmented estimation error is proposed.After segmenting each variable of the time series,this method replaces the original time series data sub-segment by calculating the quadratic power of the estimated derivative difference between the time series feature sub-segments,and establishes the feature matrix.Then corresponding weights are given accor ding to the actual usage of the characteristic sub-segments of time series.Finally,the dynamic time warping method is used to calculate whether the distance between time series is within the threshold value to judge the similarity between time series.This method solves the problems of unreasonable matching and ignoring the relationship between internal adjacent data points in the time series similarity measurement.Secondly,a dynamic clustering similarity measurement method based on improved SAX is proposed.This method expands on the basis of the SAX method that only uses the average points of the sub-sequences to replace the overall time series information,and adds characteristic information such as extreme points and boundary points to replace the original sub-sequence average symbols.On this basis,the distance measurement is performed by the dynamic clustering method without artificially setting the number of clusters to make the similarity measurement result better.This method solves the problem of not fully reflecting the changing trend of the sequence and easy to lose characteristic information when using the average value to symbolize the time series.At the same time,it solves the problem that the number of clusters needs to be preset when the traditional clustering method is used for distance measurement.Finally,the above methods are analyzed experimentally,and the experimental results show that the proposed method has good performance and can achieve an ideal similarity measurement effect.