Research On Time Series Homogeneous Relationship Discovery Based On Trend Primitive

The dissertation focuses on time series similarity measurement on structural similarity and time series homogeneous relationship mining problem.Major research results are as follows:(1)Traditional time series similarity measurement problems are transformed to distance measurement problems,which cannot reflect the essence of time series structural similarity and the change processes of the time series trends are not considered.On the distance measurement method between the continuous curve segments,the discrete time series datas are fitted to continuous curves.A judging theorem called homogeneous trend primitive is proposed and proved,which made a research foundation for the follow-up work.(2)Based on the theorem of homogeneous trend primitive,the related concepts of homogeneous subsequence are defined.To solve the homogeneous subsequence discovery problem,an algorithm on homogeneous trend primitive is proposed.The experiments about the effects of related parameters on experimental results verify the correctness of theoretical analysis.(3)For the time series trend homogeneity measurement problem,an idea based on longest homogeneous subsequence discovery is constructed.Two time series trend homogeneity measurement algorithms are proposed,which are called THMSC(time series Trend Homogeneity Measurement based on Spectral Clustering)and THMDC(time series Trend Homogeneity Measurement on Density-based Clustering)respectively.The experimental results show that the time series trend homogeneity measurement results are basically consistent on the two algorithms.(4)For the problem of maximal frequent itemsets mining in temporal association rules discovery process on frequent homogeneous trend primitive classes,an algorithm based on O-Apriori is proposed.Compared with Apriori algorithm which has to scan the database for every candidate itemset,the algorithm improves the efficiency with only once scanning the database.