Research On Similarity Of Time Series Based On Shape

Time series data are widely used in the fields of economy,finance,meteorolbgy and so on.The study of time series similarity has profound theoretical significance and important practical value.In the research of time series data mining,many algorithms are based on some similarity measure.The feature representation method of time series has an important influence on the result of similarity measurement.In this paper,the following research work has been done on the basis of the analysis of the classical time series similarity measurement algorithm and the feature representation method:Firstly summarizes the research background and significance of time series data mining,points out the research status of time series feature representation and similarity measurement.Then,the most representative methods of time series pattern representation and similarity measurement are introduced in detail Finally,two improved algorithms for similarity measurement of time series are presented.The time series itself has the characteristics of high dimension and high noise,so it is necessary to carry on the feature representation of the sequence before the similarity measure is carried out.In view of the existing feature representation methods have limitation in trend extraction and dimension reduction,and lead to the difficulty of accurately and efficiently extracting the trend of sequence shape,a new trend-based slope distance similarity measure algorithm is proposed.The algorithm combine empirical mode decomposition method with the piecewise linear representation method.The empirical mode decomposition method has the advantage of filtering noise,firstly,use it to extract shape trends,and then use the piecewise linear representation method to segment the trend sequence,the results of feature representation are optimized.On this basis,the slope distance is improved by combining the sequence pattern,which overcomes the problem that the slope distance method does not take into account the pattern difference between different time series and lead to the problem of error exists in similarity measurement results.The method of time series similarity measurement based on three element fluctuation patterns is not very detailed,cannot reflect the specific trend changes in the sequence,a new similarity measure algorithm based on shape pattern is proposed The algorithm firstly use piecewise linear representation method to represent the time series,on the basis of this,the method of dividing the seven-element shape pattern is given.According to the slope value of the sequence at different time periods,the shape pattern of the segmentation can be determined,and different shape patterns of the sequences can be represented by different numbers.Thus,the time series are transformed into special string sequences,and the longest common subsequence method is used to calculate the distance between the string sequences.Theoretical analysis and simulation results show that the two methods proposed in this paper can improve the accuracy of the results in the similarity measure of time series,and has the characteristics of good stability and is not sensitive to noise and translation.