基于概率模型的数据流预测查询算法  

Predictive Queries Algorithm Based on Probability Model over Data Streams

作　　者：李国徽[1] 陈辉[1] 杨兵[1] 向军[1] 陈刚[1] 

机构地区：[1]华中科技大学计算机学院,武汉430074

出　　处：《计算机科学》2008年第4期66-69,共4页Computer Science

基　　金：青年国家自然科学基金(No60203017);国家教育部博士点基金;湖北省杰出人才基金支持

摘　　要：挖掘在线数据流的变化趋势并预测未来时间窗口上的可能值,可以为许多时间敏感的应用提供重要决策支持。通过将数量可能无限的流数据元素映射到离散的且数量有限的流数据状态空间,不断变化的流数据变化趋势可以模拟成连续的流数据状态变化的过程,进而在很小的时间与空间代价下,数据流状态变迁的趋势动态存储在状态变迁图中。通过分析状态变迁图中的流数据变迁的统计规律,数据流上未来时刻的可能值可以应用马尔可夫模型在线连续预测。Recently, data stream has widely appeared in many applications. Mining the evolving tendency and forecasting the data values in the future time windows of streams can provide important support for the future decision in many time-sensitive applications. For example, by using predictive queries in sensor networks for all kinds of monitoring, observers can forecast future values to detect abnormal events. By mapping the stream data into the stream state space, a continuously changing tendency of an online data stream can be modeled as a state transition process. After studying the history trend of the state transitions which are kept into state transition diagraph, the values in the next m time windows of a data stream can be forecasted efficiently. Extensive simulation experiments are conducted and show that the efficiency and precision of the proposed method are better than that of the existing analogous algorithms.

关 键 词：数据流 流数据挖掘 数据预测 马尔可夫链 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP393[自动化与计算机技术—计算机科学与技术]