System identification plays an important role in the development of process simulators and controllers. The ability to determine correctly the model parameters directly affects the model quality and, therefore, the model based controller performance. This work details the development of a system identification approach and its computational implementation based on sequential quadratic programming (SQP) in which first and second order linear systems, represented in state-space, are identified from simulated and from real industrial process data. Both single-input single-output and multivariable processes are considered. The resulting optimization problem may become not trivial to solve as one of the examples illustrates. It is shown how a rescaling of the decision variables or the use of a priori process knowledge may be used in order to overcome the difficulties and to improve the quality of the results.
Digital Object Identifier (DOI)
S. R. Brásio, Ana; Romanenko, Andrey; and C. P. Fernandes, Natércia
"Using Sequential Quadratic Programming for System Identification,"
Applied Mathematics & Information Sciences: Vol. 09
, Article 3.
Available at: https://dc.naturalspublishing.com/amis/vol09/iss1/3