钟差预报AR模型异常值探测的Bayes方法

Bayesian Outlier Detection for Autoregressive Model of Clock Error Prediction

主要研究了时间序列AR模型的异常值探测问题。首先在一定的限制条件下,将AR模型的异常值探测问题转化为线性回归模型的异常值探测问题;并在正态-Gamma先验条件下,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。然后运用Bayes方法对异常值进行了估算。最后通过卫星钟差实测数据计算,比较了模型修正前后预报的情况,验证了新方法的有效性。

Outliers detection method of autoregressive model was mainly discussed.Giving some special restriction conditions,the time series outliers positioning can be transformed into the problem of outliers positioning in linear regression model.Under Normal-Gamma priori distribution,the posterior probability of outliers event based on level shift model and variance inflation model can be calculated firstly.Then the errors of outliers can be estimated using Bayesian method.At last,an experiment of real clock error data demonstrated the validity of the new method.