基于时间序列分析的在线离群点校正方法

Online outlier correction method based on time series analysis

随着合成氨项目的大型化和自动化,现场远传仪表数量越来越多,部分仪表因自身的故障导致测量数据失常,这部分检测的数据称之为离群点。在软测量模型仿真过程中,离群点对建模精度产生较大影响。本文通过引用时间序列校正算法来对检测出来的离群点进行校正,以氨合成项目中氨净值作为数据进行仿真研究,表明时间序列分析能很好的对离群点进行校正且能有效的提高模型的预测精度。

With the large-scale and automation of ammonia synthesis project,the number of remote sensing instruments on site is increasing,some instruments may experience measurement data abnormalities due to their own malfunctions,and the detected data is called outliers.In the process of soft measurement model simulation,the outliers have a greater impact on the modeling accuracy.In this article,the time series correction algorithm is used to correct the detected outliers,and the net ammonia value of ammonia synthesis project is used as the data for the simulation study,which shows that the time series analysis can correct the outliers very well and improve the prediction accuracy of the model effectively.