Research And Application On Interval Time Series Clustering Based On DTW

With the development of human society and the advancement of science and technology,time series data is widely used in finance,medical and industrial fields.It has important theoretical and application value for the study of time series data.Because time series data is widely used in many fields,such as finance,medical treatment and industry,the research on time series data has important theoretical value and application value.Interval time series data is a generalization of ordinary time series data,which is sequence data composed of interval data arranged in order of time points.Interval time series data can not only represent the specific values observed at each time point,but also include the range of data changes at time points.Because interval time series data has the characteristics of high dimension,complexity and mass,researchers directly mining data on the original data will result in high time complexity and space complexity,and also lead to the credibility of research conclusions.Therefore,it is of great research significance to explore how to more effectively mine interval time series data.The specific research work in this paper is as follows:First,A feature representation method for interval time series data is proposed.This paper takes interval time series data as the research object,and proposes Lq trend filtering-eometrical barycenter and exterior radius-based transformation(Lq-BR)for its characteristics of complex data structure and huge data volume.The algorithm using Lq trend filtering in penalty function 0<q<1 will be within the scope of intrinsic trend of time series for the characteristics of the piecewise linear form,respectively for upper boundary and lower boundary of the interval time series data of time series on the application trend of Lq filtering and piecewise linear representation.,and then the triangle is used to fill the band region formed by interval time series.The interval time series are represented by the coordinates of the center of gravity and the outer radius of the triangle,and finally the interval time series is converted into a triples series.This method can not only effectively reduce the dimension of the interval time series data,but also fully consider the time axis information of the interval time series and maximize the retention of the data information of the original interval time series.Second,Construct an interval time series data clustering algorithm model based on dynamic time warping.Considering that Dynamic Time Warping(DTW)algorithm is a similarity measurement method that can realize asynchronous data matching and has strong robustness to noise and outliers of time series data.In this paper,combining the advantages of Lq-BR feature representation and DTW,Lq-BRDTW distance measurement algorithm is proposed and applied to clustering algorithm.Experiments on the UCR time series classification archive dataset prove the effectiveness and efficiency of the proposed algorithm.Third,Construct a stock recommendation prototype system.Apply the Lq-BRDTW based clustering algorithm proposed in this paper to the interval time series data composed of the highest price and the lowest price of stocks,recommending less relevant stocks to users,and helping users avoid investment risks.