基于ARMA模型的联机时间序列数据分割算法  

An On-Line Algorithm for Segmenting Time Series Based on ARMA Model

作　　者：黄超[1] 朱扬勇[1] 

机构地区：[1]复旦大学计算机与信息技术系,上海200433

出　　处：《模式识别与人工智能》2005年第2期129-134,共6页Pattern Recognition and Artificial Intelligence

基　　金：国家863高科技研究发展计划(No.2001AA113181);上海市科委科技攻关计划(No.045115003)资助项目

摘　　要：自回归移动平均(ARMA)模型近年来被广泛用于时间序列数据的预测、聚类以及相似性查找等。由于现实中的时间序列数据其底层生成机制与结构经常动态发生变化,因而对跨越较长时期的数据建立一个单一静态的ARMA模型并不合适。本文提出一种联机分割算法,首先对数据建立动态的ARMA模型,然后根据模型的预测信息与历史数据的特征信息,判断是否适合继续使用该模型描述后续数据,或者需要对数据进行分割,从而逐段建立ARMA模型。算法能够处理持续数据流,对仿真数据与实际数据的试验结果表明,本文所提出的算法是实用有效的。Autoregressive Moving Average (ARMA) model has been widely applied to time series forecasting, classification, clustering and similarity searching recently. Since the underlying data generating process and structure of most real--world time series often change dynamically, it is not suitable to construct a single and static model for time series that span a long time. This paper brings forward an on--line algorithm. Firstly, this algorithm constructs a dynamic ARMA model for time series, then judges whether this model is suitable for the following data or we should segment the data and construct models piecewise by using forecasting information and the characteristics of history data. It can process continuous data stream. The experimens conducted on simulated and real--world datasets show the algorithm is practical and effective.

关 键 词：时间序列 自回归移动平均模型 分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]