Modeling uncertain processes with interval random vector functional-link networks. (September 2020)

• Record Type: Journal Article
• Authors: Guan, Shouping; Cui, Zhouying
• Is Part Of: Journal of process control. Volume 93(2020)
• Page Start: 43
• Page End: 52
• Publication Date: 2020-09
• Subjects: Interval neural network (INN) -- Interval random vector functional-link network (IRVFLN) -- Interval process modeling -- Glutamic acid fermentation process -- Unknown but bounded (UBB) error; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2020.07.008
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract: This paper presents a new approach to building an interval model for an industrial process with uncertainty that employs an interval neural network (INN), which can solve problems such as model structure demands and complexity limitations in the conventional unknown but bounded (UBB) errors method. A new architecture for an interval random vector functional-link network (IRVFLN) and its learning algorithm with penalty factors are proposed, to solve the problems such as the local minima, slow convergence, and very poor sensitivity to learning rate settings in the interval feed-forward neural networks with error back-propagation (IBPNNs). As an application case study, the IRVFLN is used to model the glutamic acid fermentation process under the condition of bounded-error data, and the test results indicate that the accuracy of the IRVFLN model meets the manufacturing requirements. The comparison is performed with IBPNN, and the results demonstrate that the proposed network outperforms IBPNN both on effectiveness and efficiency. Also, a comparison is given with a crisp (point-valued) approach using RVFLN, and the results show that the crisp approach is less reliable when existing uncertainties in measuring or process. Highlights: A novel strategy to model an industrial process is proposed. A new interval network of IRVFLN is presented. The IRVFLN is applied to model a complex industrial process.