Mining Dynamic Association Rules From Multiple Time-series Data Based On Sliding Window

With the rapid development of economy, there will have a lot of data in the fields of industrial production, financial services, electronic commerce, satellite remote sensing, sensor network and so on. These data often have time label. It is an important research direction to excavate the correlation of multiple time series flow in a certain domain. Because time series stream has the characteristics of mass, real-time and continuity, the data and knowledge will change with time, Traditional association rules mining method is difficult to carry out effective mining.At present, there have been some research on data stream association rules mining, but many of them have been symbolized by the data stream or that the data itself has represented a pattern that does not require preprocessing to extract patterns. The further study of multivariate time series association rule mining is less, and a lot of mining method use the same length to extract patterns, and the rules of the mining are all the same time length nor does it take into account the more interesting rules in the new data.In this paper, a sliding window is used to limit the time series data, Mining dynamic association rules namely the rules dynamically change with the sliding window moves. Because the time series data is continuous, it needs to be preprocessed to extract meta data sets to form transactions. In the process of preprocessing, the data is linearized. The linearized data is cut so that the multiple series only have one pattern at the same period of time, and then the similar line segments are clustered together. So the multiple time series data is symbolized, the association rules can be extracted. The algorithm of this paper is based on the sliding window. In the sliding window maintenance a summary structures SWIU-tree(Incremental Updating tree based on Sliding Window) a global storage for scanned transactions, by pruning strategies to remove less frequent pattern and expired mode of SWIU-tree. At the same time, take the strategy of counting and decay on the different basic window of the sliding window, reducing the impact of historical items.On the actual thermal power plant data and Stock data, by comparing the existing algorithm with the SWIU-tree algorithm, the proposed algorithm shows the effectiveness of the proposed algorithm, which can quickly and accurately mining association rules of multiple time series data.