The Study Of Data Stream Clustering Based On Grid And Fast Particle Swarm Optimization

In recent years, the technology of data acquisition has become more and more intelligent,automated, large amounts of data are produced in a very short period of time. These data are different from the traditional data. They are called data stream possessing the features such as timeliness, continuity and continuation. The traditional technique of clustering analysis can not be applied to the data stream directly so it is very difficult to cluster data stream. Particle Swarm Optimization algorithm belongs to the category of Swarm Intelligence. It emphasizes the interaction between individuals and has good self-adaptability, self-organization and robustness characteristics. At the same time, the features of data stream require clustering algorithms to have these features. Furthermore, some of the current data stream clustering algorithms have the problems of low efficiency, poor adaptability etc... Therefore, it will not only be a practical method but also have good development prospects to apply swarm intelligence technology to data clustering analysis.This paper, through the research on data stream clustering algorithm based on grid, combines the algorithm with fast Particle Swarm Optimization. A more efficient algorithm——GFPSOS(Grid and Fast Particle Swarm Optimization on Stream) is proposed. The algorithm uses the two stages structure of CluStream algorithm, which processes data in the online stage, then clusters data and returns the result to user in the offline stage. However, the offline stage of CluStream algorithm use the method based on distance, which makes it difficult to find non-spherical clusters. So in this paper, we use technologies of grid to remove low density grids dynamically to make it possible to find non-spherical clusters. At the same time, we use the fast Particle Swarm Optimization which adds pattern reduction operation and multi-start operation to original Particle Swarm Optimization to improve the efficiency.Through the experiments and analysis, GFPSOS algorithm can deal with non-spherical clusters, compared with the current popular data stream clustering algorithms, it also has better precision and higher efficiency of clustering.