Inferential control with the aid of modified QPLS-based soft sensor for an industrial FCCU fractionator

Abstract

A modified quadratic partial least squares (MQPLS) algorithm based on non-linear constrained programming is proposed, in which a sequential unconstrained minimisation technique is employed to calculate the outer input weights and the parameters of inner relationship. Other existing quadratic partial least squares (QPLS) algorithms are also reviewed and compared with the proposed MQPLS in the applications to two datasets, one being an artificial dataset and the other being the real data from an industrial fluidised catalytic cracking unit (FCCU) main fractionator. It is shown that the MQPLS not only can explain better the underlying variability of the data but also achieves improved modelling and predictive performance over the existing QPLS algorithms. An inferential control system is implemented on the distributed control system for an industrial FCCU main fractionator, in which the soft-sensor is built based on the MQPLS algorithm to estimate the diesel oil solidifying point online and the controller is established via a constrained dynamic matrix control algorithm. Experimental results obtained demonstrate that the inferential control system with the aid of the MQPLS soft sensor works much better than the original tray temperature control system and it realises well the bounder control of diesel oil solidifying point.