Premium
Variable structure NARX models: Application to dissolved‐oxygen bioprocess
Author(s) -
Sargantanis Ioannis G.,
Karim M. N.
Publication year - 1999
Publication title -
aiche journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.958
H-Index - 167
eISSN - 1547-5905
pISSN - 0001-1541
DOI - 10.1002/aic.690450920
Subject(s) - nonlinear autoregressive exogenous model , autoregressive model , control theory (sociology) , process (computing) , model selection , bioprocess , system identification , continuous stirred tank reactor , selection (genetic algorithm) , computer science , mathematics , engineering , statistics , artificial intelligence , data modeling , control (management) , database , chemical engineering , operating system
Selecting the best structure of nonlinear autoregressive models with exogeneous inputs (NARX) is important for good parameter estimation and control. If the process is also time varying, the best structure selection procedure should be done in an on‐line fashion. Simultaneous model selection and parameter estimation can be achieved by employing the Givens forward selection with exponential windowing (GFEX) algorithm. An on‐line implementation yields a variable‐structure NARX‐model identification, which can be used for control of a dissolved‐oxygen (DO) process. The results of an adaptive DO control scheme, which uses a fixed model structure and was applied to a batch Bacillus subtilis fermentation, were used in this study. Time‐dependent patterns of selected terms were observed that roughly corresponded to different fermentation stages. Fitting and prediction capabilities for the adaptively selected NARX model were compared with fixed ARX and NARX models. The effect of the forgetting factor and the model order was investigated. Performance is satisfactory for short‐step‐ahead prediction, and avoiding overparametrization can stabilize the long‐step‐ahead prediction.