基于共同趋势模型的非平稳过程在线监控

Online non-stationary process monitoring by common trends model

对于非平稳过程监控,传统的基于数据平稳假设的多元统计过程控制方法是不适用的。针对上述问题,提出了一种基于共同趋势模型的非平稳过程监控方法。共同趋势模型从存在协整关系的非平稳多元变量中辨识出共同因子,将各非平稳过程变量分解成非平稳的共同趋势成分与平稳成分之和的形式。不同于现有的基于协整模型的非平稳过程监控方法,共同趋势模型能够获取各非平稳变量中的平稳成分,消除非平稳共同因子的影响并体现变量间全部的动态均衡关系。将对非平稳过程的监控变为应用共同趋势模型,分解得到各非平稳过程变量中的平稳成分,然后应用传统的多元统计方法,估计平稳成分的统计量及相应的控制限进行监测。石油蒸馏过程监控的实例研究结果表明,所提出的方法比基于协整新息变量的方法具有更可靠的监控效果。

Non-stationary process monitoring based on common trends model was proposed, because conventional multivariate statistical process control methods with stationary data assumption were inapplicable to non-stationary process monitoring. The new common trends model was capable of identifying common factors from co-integrated non-stationary multiple variables and decomposing each non-stationary process variable into summation of a non-stationary common trends component and a stationary counterpart. Contrary to existing non-stationary process monitoring technique from cointegration model, the common trends model captured the stationary component of each non-stationary process variable, eliminated effects of non-stationary common factors and unveiled overall dynamic equilibrium relationships among variables. Hence, non-stationary process monitoring was transformed to an application of common trends model, which involved obtaining stationary component of each process variable, creating estimation for the stationary components by conventional multivariate statistical methods and setting up monitoring on corresponding control limits. A case study of monitoring petroleum distillation process showed that the proposed approach possessed more reliable process monitoring performance than the method of cointegration model.