Research on data reconciliation based on generalized T distribution with historical data

详细信息    查看全文

• 作者：Shengxi WuAuthor Vitae ; Qiang YeAuthor Vitae ; Cheng ChenAuthor Vitae ; Xingsheng Gu ; ^{xsgu@ecust.edu.cn" class="auth_mail" title="E-mail the corresponding author}Author Vitae
• 关键词：Data reconciliation ; Maximum likelihood estimation ; Generalized T distribution ; Historical data ; PSO algorithm
• 刊名：Neurocomputing
• 出版年：2016
• 出版时间：29 January 2016
• 年：2016
• 卷：175
• 期：part_PA
• 页码：808-815
• 全文大小：780 K

文摘

In the most of previous data reconciliation(DR) studies, process data were conventionally characterized by normal Gaussian distribution, so the optimality/validity of DR estimator is implicitly based on a main assumption that errors follow normal Gaussian distribution. When this assumption is not satisfied, conventional data reconciliation approaches will become unavailable. However, normal distribution usually does not exist in real chemical engineering practice, as it is hard to ensure the normality even for high-quality measurements. So it is necessary to propose a new DR method which can accommodate more variety of measurement error distribution. In this paper, generalized T distribution is applied to accommodate measurement error distribution, meanwhile, historical data is introduced to estimate the objective function parameters by using Particle Swarm Optimization (PSO) algorithm. A novel robust data reconciliation method is proposed based on GT distribution and historical data, at the same time, its robustness characteristics are investigated. The new method is demonstrated on a steam-metering system for a methanol synthesis unit. Based on the comparison with other DR methods, the novel robust DR method can effectively improve the reliability of reconciled data even when errors do not follow normal distribution.