Autoregressive exogenous input modelling for lipase catalysed esterification process

One of the effective ways to enhance the productivity of esters is by controlling the process at its optimum conditions. The crucial task of model based control is the development of an appropriate process model. Model requires to be less complex but with adequately accurate. The development of a suitable empirical model is imperative to ensure the best control performance for model based controller. In this works, a linear Autoregressive with Exogenous input (ARX) and a Nonlinear Autoregressive with exogenous input (NARX) models were developed and compared for lipase catalysed esterification batch process. The identification and validation data for the ARX and NARX models were generated from the first principle model. The inputs data included were jacket flowrate, air flowrate and jacket temperature. Meanwhile, the outputs data considered were temperature and water activity. The empirical model developed was determined by varying the number of inputs/outputs model order and the model parameters involved were estimated using the Kalman filter method in the Recursive Least Squares Estimation (RLSE) function block in MATLAB (R)/Simulink. The accuracy of the model was the main indicator for selection. The values of R-2 for the ARX and the NARX models for the reactor temperature were 70.80% and 98.06%, respectively, while for the water activity the values of R-2 for the best ARX and NARX models were 63.46% and 99.04%, respectively. The results show that the NARX with low order models fit the real data very well if compared to the ARX models. (C) 2020 International Association for Mathematics and Computers in Simulation (IMACS). Published by Elsevier B.V. All rights reserved.