The Study Of Indexing And Retrieval Algorithm For Symbolic Multi-Dimensional Time-series

Data mining techniques that focus on multi-dimensional time series attracts more and more researchers,and the indexing and retrieval methods is the one of the most important part,with the increasing of the dimension of data,data mining algorithm gets more and more complicated,many researchers suggested that it is important to using symbolization method to solve the problem.For the purpose of symbolization,many algorithms well known is designed for one-dimensional sequence,which are not suitable for multi-dimensional data.Advances a method of multi-dimensional sequence symbolization based on multi-level k-means clustering.Granularity of symbolization is determined by the max square errorτ_max.For the purpose of indexing and retrieval of symbolization sequence,advances a method based on inverted index data structure.First conversion coarse filtration problem to t-threshold set problem.Then advances a complete partition method based on heap data structure,which improve the efficiency of algorithm from incomplete partition method.Using longest common subsequence(LCSS) as the measure,which improves the original LCSS algorithm by defineρ_min as the minimal matching rate and using characteristic of inverted index model.The method allows an elastic shifting of the time axis.Define the match of multi-dimensional symbol,so it can be used on indexing and retrieval multi-dimensional sequence.At last,using the experiment on indexing human sports data to show that high efficiency of the algorithm and also the correctness,even when added noise data.