Soft sensor model for monitoring and online control based on a dynamic model and local instrumental variable technique

The aim of this paper is the design of two data-based soft sensors for accurate prediction of isopropyl benzene concentration in an industrial distillation column. The first soft sensor is based on the state-dependent-parameter model and a local instrumental variable (LIV) method relying on the static data. The main novelty of this work is focused on the second soft sensor, which is introduced to compensate the time lag ignorance in the first proposed soft sensor. A dynamic model is considered between predicted values of LIV-based soft sensor and simulated concentration by Aspen. Their performances are evaluated by offline mode and industrial and simulated data and also, by online control structure with a proportional-integralplus controller. The results of non-parametric models show a very low error percentage and supreme agreement with prediction quality from the rigorous model compared with other models.