Research On Ranking Result Based Fast Subsequence Similarity Search

Time-series similarity search is finding similar sequence or subsequence in time series databases with a given sequence, which is a new and significant analysing method of time-series data and has wide application prospects in many domains. This paper presents a comprehensive analysis that the current situation of the domestic and international research about time-series similarity search. On these bases, subsequence similarity research needs further research. The detailed content can be illustrated as follows.Firstly, already proposed dimensionality reduction cause unboundedness of searching space, so subsequence similarity search algorithm can not effectively filter the point in the index. Thus, a new non-linear dimensionality reduction is used to resolve the problem. Then, early abandon technique is introduced to reduce redundant computation. A ranked subsequence similarity search algorithm is proposed to realize ranking query result.Secondly, in the face of the data sequence changing and the query parameter updating dynamically, using incremental array record the computing distances of last query, through which the algorithm can acquire a optimal parameter of this query in order to reduce redundant computation directly using subsequence similarity search algorithm. When the data sequence increases, the data sequence decreases, or the query parameter updates, incremental ranked subsequence similarity search algorithm is proposed respectively.Thridly, in order to prevent privacy disclosure in the course of ranked subsequence similarity search, a preventing privacy model is proposed which hide time series data through Mean Standard deviation transformation and protect the mean of time series through label replacement strategy. A secure distance computation protocol is proposed to compute the distance between query sequence and corresponding subsequence securely. A detailed analysis of security model includes correctness, security, accuracy and computational cost. On these bases, a ranked subsequence similarity search algorithm based on privacy preserving is proposed.Finally, the algorithms in this paper are validated. The figures of result are given and the results are analysed and compared.