基于残差统计的时间序列加性离群点检测算法研究

Residuals statistics-based additive outlier detection algorithm for time series

针对时间序列,提出了一种基于残差统计的加性离群点检测算法,利用AR模型对时间序列进行前向与后向拟合;采用了数据相对变化率判别法减少离群点对拟合的影响;根据假设检验原理,以高斯分布统计检验对残差进行统计分析并最终确定离群点。仿真结果表明,该方法对离群点检测有较高的准确性。

We propose a residuals statistics-based additive outlier detection algorithm for one-dimensional time series, The basic idea is using time series AR model for forward and backward fitting. In order to reduce the influence of outlier, we use data＇s relative change rate to preliminary judge the outlier. According to hypothesis testing theory and Gauss distribution statistic testing, we find out the outliers. The simulation results show that the this method has good performance on outlier detection.