Multivariable statistical modelling and its application

The work described in this thesis isconcerned with the study of statistical identification ofdiscrete, dynamic, multi-input and multi-output CNIMQ)linear models for complex industrial processes.Contributions that have been made in this area include:1. The development of a MIMO estimation procedure, that iscomputationally quite simple, to estimate theparameters in the discrete MIMO model. The resulting parametersare shown to be asymptotically equivalent tothe maximum likelihood estimators.2. The development of a suitable methodology so that theestimation procedure can give results in the practicalidentification case where the data is not always ideal.The ideas and methodology developed inthe thesis are tested on practical data collected from areal plant. The data has been logged from a 50QMW boiler—turbine unit at the Electricity Commission of New SouthWales's Liddell Power Station. Good models have beenobtained using the methodology developed, and the modelsgive output responses that are in good agreement with theplant output responses.In this thesis, we have successfully demonstratedthat it is possible, for control purposes, toobtain a useful boiler model from input-output data recordsof short length using the statistical identificationapproach. The methodology developed is equally applicablefor identifying models for other processes