AR序列异常值探测的Bayes方法在卫星钟差预报中的应用  

The Application of Bayesian Method in Autoregressive Model Outliers Detecting of Satellite Clock Error Prediction

作　　者：李涛[1] 衡广辉[1] 归庆明[1] 

机构地区：[1]信息工程大学理学院,河南郑州450001

出　　处：《全球定位系统》2010年第4期15-20,30,共7页Gnss World of China

基　　金：国家自然科学基金项目(40974009);河南省基础与前沿技术研究计划项目(082300410240);郑州市科技计划攻关项目(0910SGYG21198)

摘　　要：AR模型中若含有异常值,会使传统的建模、估计及检验方法陷人困境,从而不能准确地预测和控制。本文在无信息先验条件下,结合观测信息,计算了基于均值漂移模型和方差膨胀模型异常值事件发生的后验概率,提出了无信息先验下自回归模型中异常值探测的Bayes方法并对异常值进行了估算。该方法能将异常值准确地探测和估算出来,借此修正模型,可提高预测的准确性。最后,做了钟差实测数据计算,比较了模型修正前后预报的情况,验证了该方法的有效性。We need to establish proper satellite clock error model to prediction. The autoregressive model of satellite clock error is an efficient modeling method. But, when there are outliers in the clock error, it will be difficulty for the traditional modeling method, estimation method and testing method. Then the prediction will be inaccurate. So, it is important to find proper outliers detection method. This paper will mainly discuss the outliers de- tection of autoregressive model. Given noninformative prior, combined with observed infor- mation, we calculate the posterior probability of outliers event based on level shift model and variance inflation model, then put forward Bayesian method of outliers detecting in autoregressive model and estimate the outliers. The method can detect and estimate the outliers exactly. In virtue of？ this, we can modify the model and promote the prediction accuracy. At last, we make an experiment of real clock error data, compared with the prediction results on unmodified and modified. The new method tested to be efficient.

关 键 词：卫星钟差 预报 AR模型 异常值 BAYES方法 

分 类 号：P228[天文地球—大地测量学与测量工程]