基于动态滑动窗口的改进数据流聚类算法

摘要

An optimization algorithm DCluStream was proposed which processed data over sliding window. The method adopted online-offline clustering framework of CluStream.The real time of the data object coming and out of sliding window was introduced into the characteristics of the cluster,adjusting the win-dow size reasonably in the limited memory resources environment.Using the time decay mechanism on his-torical data could reduce the impact of new data object,which could get better clustering results.The exper-imental results showed that compared with the algorthm CluStream,data processing efficiency of the algo-rithm was relatively higher with saving memory.%提出一种采用滑动窗口处理数据的优化算法DCluStream．该方法基于CluStream算法双层框架思想，在聚类特征中引入数据流入和流出滑动窗口的实际时间，动态调整窗口大小以适应有限内存；对历史数据通过时间衰减机制来降低它对新数据对象的影响，使聚类效果更好．实验结果表明，与CluStream相比，本算法处理数据的效率更高且相对节约内存．