The Index Of Time-series Based On Orthogonal Transformation

Time series similarity measure and indexing are the key and foundation for data mining. In this paper, we first have a more systematic literature summary on data mining technology and analyze the existing relevant technologies,then describe a few basic problems, such as dimension reduction of high dimensional data, similarity measure and indexing on time series. We talk about the advantage and disadvantage on the problem and have a more in-depth study. Finally, we propose the hierarchical tree in time-series databases and the retrieval of orthogonal transformation of time series, at the same time we verified it through the experiments.Time series data has the following characteristics:mass, complexity and nosie disturbance. If we search on the original series directly, they will appear large computation; the accuracy and reliability are difficult to be guaranteed. At present, the representation of time series, domestic and foreign researchers haveproposed many algorithms. The algorithm extracts the characteristic value or rectangle of the time series, implements data dimension reduction and then extract the characteristic value or rectangle to bulid efficient index structure, so that the time series query is increased greatly.The hierarchical tree in time-series databases is to handle high-dimensional time-series with the lower time complexity. We introduce a new index structure-Hierarchical Tree,for efficient time-series retrieval and similarity query. We divide time-series into segments, and build different level R*-Trees. We introduce the insert and query algorithm.And we prove no false dismissals for the algorithms. Extensive experiments reveal that the retrieval speed of Hierarchical tree is superior to the traditional retrieval method if the time-series databases are high-dimensional and similarity.This paper reviews the time series and indexing techniques briefly. Though the introduction on the R-tree, R*-tree and X-tree, we can present a complete and orderly development of the retrieval structure diagram, which in turn are based on the tree for improvement. It has a higher efficiency and applicability. In this paper we present a similarity measuring and matching method of orthogonal polynomial based represented time series.We establish a layered search tree index according to the value extracted from the orthogonal polynomials.The hierarchical tree can lower the time complexity of the establishing of indexing structure, the proposed algorithm is still able to matain good spatial and temporal complexity. At the same time, it proved the indexing of the orthogonal transformed time series have non-false dismissal nature in the sense of DTW similarity measure.The contrast experiments against the time series similarity inquiry based on the Discrete Fourier Transform and Discrete Wavelet Transform, we can obtain higher query efficiency with lower dimension of index structure.