Application of artificial neural network‐based generic model control to multivariable processes

Abstract

AbstractThe present study focuses on the application of a purely data‐driven artificial neural network (ANN)‐based generic model controller (GMC) for multiple‐input and multiple‐output systems with relative degree one or more. The approach is based on development of an ANN model relating the time‐derivatives of the outputs of the order equal to their relative degree with the past measured values of input and output variables and formulation of a constrained optimization problem to satisfy GMC performance objectives. Two variants of a multivariable semi‐batch reactor (SBR) control problem for esterification reaction of maleic anhydride with hexanol to form hexylmonoester of maleic acid are considered without and with consideration of coolant dynamics (SBR1 and SBR2, respectively). Hexanol concentration and reactor temperature are considered as the outputs, and the coolant and feed flow rates are considered as the manipulated inputs. The present study illustrates that the proposed multivariable ANN‐based GMC exhibits setpoint tracking and disturbance rejection performance similar to the exact model‐based GMC for both the reactors, and the ANN‐based GMC input profiles are smoother and closer to the nominal input profiles compared with GMC controller, although no process information is used other than the output measurements in the proposed approach. © 2017 Curtin University and John Wiley & Sons, Ltd.