Multivariable dynamic modeling for molten iron quality using online sequential random vector functional-link networks with self-feedback connections

This paper presents a data-driven dynamic modeling method for the multivariate prediction of molten iron quality (MIQ) in a blast furnace (BF) using online sequential random vector functional-link networks (OS-RVFLNs) with the help of principal component analysis (PCA). At first, a data-driven PCA is employed to identify the most influential components from multitudinous factors that affect MIQ so as to reduce the model dimension. Secondly, a dynamic OS-RVFLNs modeling technology with fast learning and strong nonlinear mapping capability is proposed by applying the output self-feedback structure to the traditional OS-RVFLNs. Since it has been shown that such a dynamic modeling method has the ability to store and handle input-output data at different time scales, the dynamic OS-RVFLNs based MIQ prediction model has exhibited the potential for multivariable nonlinear mapping and the adaptability to dynamic time-varying process. Finally, some industrial experiments and comparative studies have been carried out on the 2# BF in Liuzhou Iron & Steel Group Co. of China using the proposed method, where it has been demonstrated that the constructed model produces a better modeling and estimating accuracy and has faster learning speed than other conventional MIQ modeling methods.