Trend Sequences Analysis of Temporal Data and a Subsequence Matching Algorithm

In current trend sequences nominal scale and edit distance are used to measure trend values, distance between trend sequences respectively. The analysis of this kind of trend sequences essentially belongs to the domain of character string analysis. These traditional trend sequences are called character trend sequence (CTS) in this paper. The largest problem about analysis of CTSs is to use very few indexes to depict trends of sequences which have a very large range of variety, so little information included in temporal data sequences is preserved in CTSs. To overcome demerits of traditional trend sequences' analysis in temporal data mining, two concepts which are number trend sequence (NTS) and trend sequences unwrapping are put forward. According to features of NTSs, radians which slopes correspond to are used to represent trends of line segments. Dynamic time warping double restrictions quick searching (DTW-DRQS) algorithm is designed to solve the problem of subsequence matching between NTSs. The algorithm includes three parts: DTW sequential searching, the mechanism of double restrictions and the mechanism of redundancy control. DTW sequential searching is the basic framework of the algorithm, and; the mechanism of double restrictions can accelerate the calculation process of DTW distance; the mechanism of redundancy control can eliminate redundant subsequences in the result set.