ARMAX时间序列模型异常点及异常点斑片的估计和检测  

ESTIMATION AND DETECTION OF OUTLIERS AND PATCHES IN ARMAX TIME SERIES MODELS

作　　者：陈平[1] 陈钧 

机构地区：[1]东南大学数学系,南京210096 [2]中国电器科学研究院广州威凯检测技术研究院能力验证部,广州510663

出　　处：《系统科学与数学》2010年第10期1323-1333,共11页Journal of Systems Science and Mathematical Sciences

基　　金：国家自然科学基金(10671032)资助课题

摘　　要：将通常的Gibbs抽样和自适应的Gibbs抽样算法用于带有外生变量的自回归移动平均时间序列(ARMAX)模型的Bayes分析,首先采用一些方法消除ARMAX模型中输入(外生变量)序列的影响,然后在前人工作的基础上给出了一种类似的挖掘相应时间序列中的异常点及异常点斑片的方法.说明了自适应的Gibbs抽样算法也能够有效地检测ARMAX模型中孤立的附加型异常点及异常点斑片.实际的和模拟的结果也显示这些方法可以明显减少掩盖和淹没现象的发生,这是对已有工作的推广和扩充.In this paper,the usual Gibbs sampler and adaptive Gibbs sampler are used in the Bayesian analysis of autoregressive moving average with exogenous variable（ARMAX） time series models.Firstly,some methods are used to delete the influence of input process in ARMAX model,and then the method of mining outliers and patches in time series based on the former work is given.It is shown that the adaptive Gibbs sampler is also useful in handling additive isolated outliers and outlier patches in ARMAX model.Practical and simulation studies also show that the procedure can reduce possible masking and swamping effects,and hence improve the existing methods.

关 键 词：时间序列 附加型异常点 异常点斑片 ARMAX模型 GIBBS抽样 

分 类 号：O211.61[理学—概率论与数理统计]