一种基于混沌理论的数据流连续聚集查询预测算法

A Chaos-Based Predictive Algorithm for Continuous Aggregate Queries Over Data Streams

为了有效地预测聚集查询的未来聚集值,提出了一种基于混沌理论的数据流连续聚集查询预测未来聚集值算法——CSPA算法.数据流看作是以数据到达时间为序的一个时间序列,借鉴传统时间序列分析技术探讨了连续聚集查询的未来聚集值预测问题,但由于数据流序列与传统时间序列在时间间隔和数据集的处理上存在很大差别,于是采用流滑动窗口技术加以处理.其次,针对目前数据流聚集查询预测领域已有的一些研究结果都未考虑流数据内在的复杂非线性动力学特征对预测的影响问题,该算法又利用了混沌理论中的局域预测思想解决了这一不足.实验结果表明,利用该算法进行预测具有很好的准确性.

CSPA （chaotic stream predictive algorithm） is proposed to predict efficiently the prospective aggregate values of the aggregate queries which are continuous and over data streams, based on the theory of chaos. Regarding the data stream as a time series where all the arrival times of data are arranged in order, the prediction of the prospective aggregate values of continuous aggregate queries is discussed in view of the conventional analysis of time series. However, a data stream series differs greatly from conventional time series in both time interval and data set processing, the moving window technique is therefore used for stream processing. In addition, the influence of the complex inherent nonlinear dynamic characteristics in streaming data on the prediction had not been considered in relevant earlier works. So, CSPA makes use of the idea about local prediction included in the theory of chaos to make up for the deficiency. Experimental results showed the high exactness of using the CSPA algorithm.