Research On Time Series Data Mining Of Partial Periodic Patterns

Time series data are widely existing in our lives with the development of information technology, such as stock prices, hydrological data, sale of goods and meteorological data etc. Meanwhile, the quantity of time series data has explosively increase. According to these massive historical time-series data, translate data into reliable knowledge and improve the human ability to predict the future has been closely concerned by using new technology and methods. Time series data mining aims to find out decision-making valuable information in huge data, and becomes one of the important research direction in data mining. This paper chooses the mining of time series partial periodic patterns as the main object of study, because it is more common, reserch value and significance, and it becomes a research focus in time series data mining.Firstly, this paper summarizes status of the time series data mining, and studies on some algorithm of partial periodic patterns mining in time series. At present, many potential periodic patterns will not be mining, because most algorithm are carried out under the given cycle. Aim at the defects, Weight-based POS(the events'position in time sieries) solution algorithm is proposed by combining binary code of time series, and then derives the cycle threshold formula. We can find a meaningful period of time series by the potential cycle discovery algorithm.Secondly, the paper analyzes Max-subpattern Hit-set Algorithm and chained data structure algorithm, they had improved efficiency on partial periodic patterns mining related to the Apriori-like algorithm. However, they still need to scan the time series twice and have to statistic the count of frequency count. Aim at the defects, the paper innovatively introduces concept lattice into the time series partial periodic patterns mining, and proposes a time series partial periodic patterns mining algorithm based on concept lattice. It not only reduces to one time for scaning the time series and directly reads the count of frequent patterns, but also combines with the incremental algorithm and the mining algorithm .Finally, the simulation experiment proved that the algorithm is effective.