A Study On Time Series Data Mining

Data Mining has attracted a great deal of attention under the development of information technology. After surveyed major issues in data mining, problems on time series' representation and searching are analyzed. Some algorithms and solution of association rule mining are proposed. The main ideas in this dissertation are listed as follows:1) To measure the trend similarity of time series effectively, trend representation of time series is proposed. The distance of trend series (DTS) can be calculated quickly after trend projection. DTS overcomes the problem of time series mismatch based on point distance. According to the numbers of segmentations, DTS has multi scale feature and can reflect different trend similarity of time series under various analyzing frequency.2) An enhanced algorithm, based on dual threshold value, and the conception of sub-series linear are proposed. Relative point average error is used to measure the linear degree of sub series, which produced by Bottom_Up algorithm. This method improves the accuracy of piecewise linear representation. The simulation on synthetic data and stock index is demonstrated.3) For fast sub trend sequence searching, a various steps algorithm is proposed based on analyzing the similarity of adjacent sub series distance. Compared with current methods, the result is satisfactory. After analyzed the redundancy of match sub-series, one simple solution is proposed. The simulation on synthetic data is made and the result is discussed.4) The dynamic time warping (DTW) theory is applied to actual industrial data to measure the trend similarity of various length time series, which caused by different sample time or chemical reaction time etc. The DTW distance of trend switch series is proposed to filter candidate series and improve the efficiency of similarity searching. The simulation on synthetic data is made and the result is discussed.5) By using Borland C++ Builder and Matlab, a semi system, I_Miner, is developed to mine association rule. It can realize other functions just as data cleaning, data transform, etc. Furthermore, I_Miner is applied to an actual industry database.Finally, the dissertation is concluded with a summary and prospect of future time series data mining researches...