Approach To Dynamic Pattern Discovery And Trace In Data Streams

Dynamic pattern discovery and trace is an important technology to obtain information from real-time data streams. It dynamically analyses the data streams with time stamp to discover the significative patterns and trace the variation of patterns,which is of great significance to research and practice.In this thesis, grid-based clustering method based on damped window is adopted to discover the patterns in the real-time data streams. What's more, the pattern snapshots generated in different time are stored according to the pattern storage strategy, which is used to be matched with patterns generated in current time. Then, the comparative analysis is helpful to pinpoint and describe the patterns variation, so that the pattern evolutionary process can be traced. In the implementation of the approach to dynamic pattern discovery and trace, the pattern feature tree is structured based on the structure of density dimension tree to store and restructure the patterns discovered in the data streams. Self-adaptive pruning strategy and pattern snapshot storage strategy are adopted to reduce the consumption of time and space effectively and make it more efficient.The experimental results based on synthetic dataset and real dataset demonstrate that the algorithm proposed in this thesis can effectively discover and trace the dynamic patterns in the data streams. Meanwhile, the algorithm has the advantages of low memory consumption, high processing speed and much practical use.