Research On Representation Of Time Series Based On Fitting Error And Verification Of Time Series Based On Time Difference

The information age has brought a large amount of data,a large part of which is presented in the form of time series.These data have the characteristics of high dimension and many attributes,which not only occupy a lot of computer storage space,but also need a long time to process.Therefore,it is necessary to re-represent the original time series to simplify the representation of time series.In the classification task of these time series data,some time series in same class have small similarity,and some time series in different class have large similarity.Since the original time series representation method failed to highlight the similarities of time series in same class and the differences of time series in different class,so it is necessary to re-represent the time series.In addition,in the task of time series data mining,it is one of the important tasks to judge whether a time series belongs to a certain class.Many algorithms to solve this problem are based on dynamic time warping distance.However,the classical dynamic time warping algorithm does not consider the similarity of the corresponding position(i.e.time)of each pair of matching points on each matching path of the two sequences.In view of the above two problems,this thesis proposes a time series representation method and a time series classification method(1)In this thesis,a new method of time series piecewise constant approximation is proposed.It is based on the fitting error of replacing the time series subset with a segment of constant series to partition the time series.From the first segment,greedy strategy is used to expand the length of each segment until the fitting error exceeds a certain threshold value so as to simplify the representation of time series.In addition,this thesis also clusters all elements of all the time series obtained based on this representation.According to the correlation between each cluster in these clusters and a certain time series class,more relevant clusters of the training set is selected as the representation of time series in this class of the training set so as to highlight the similarity of the time series in same class.(2)In this thesis,we improve the dynamic time warping algorithm.In the selection of the optimal matching path,we not only consider the distance between each pair of matching points in each matching path,but also focus on the similarity between the corresponding positions of each pair of matching points in the time series,and apply this algorithm to the authentication of dynamic handwritten signature.The results show that the proposed algorithm can effectively improve the accuracy of time series classification.