Research On Time Series Data Mining Algorithm Based On OLAM

In order to deal with and analyze huge data, OLAP technology and DM technology have come into being. Among them, OLAP can be used to query multi-dimensional data, DM can discover the potential and useful knowledge from data. However, in application OLAP and DM specify on their own fields, both of them have their limitation. So if we can integrate them together to develop a new DM technology based on OLAP data cube and DW, it will meet the practical demand more compatibly. OLAM technology is just the outcome of this integration. So at first this thesis will discuss related theory and research about OLAM technology detailedly, then it will introduce a kind of DM system in open-style DW integrated environment platform.Integration of data cube calculation and traditonal DM algorithm is the core of OLAM technology, this integration improves its capability mainly by using materialized views. The materialized view used in DM of timeseries is temporal view. Be aimed at a kind of temporal aggregation view, according to the pure deletion and pure insertion concept, the incremental computation of temporal views can be obtained by iterative method. Second, by using the hierarchical encoding of dimension tables, it can highly improve the efficiency of aggregation. Reasonable selection and fast maintenance of temporal view can help the DM of time series to use any time series in every granularity.Similar timeseries search is one of the important function in the DM of time series, the most difficult problem it facing is that the search space is too huge and the calculation is too complicated. This thesis provide a hierarchical method to match similar time series search. At first, this algorithm obtains rough similar series by clustering the broad moving average; Second, based on this, we can construct the similarity degree of time series trend, by using this similarity degree of time series trend, the users can do riddling secondly. At last, through calculating the distance of the residual time series, accurate similar time series are reached. Experiments show that this hierarchical approach is effective and have OLAM characters.