Multivariable process identification

This paper presents a method of multivariable system identification which utilizes the possible structures of the system to achieve a model which optimally generalizes over the available input/output data. Pseudo-observable forms are defined to illustrate the various structural forms which are possible for the system. A modified version of an associated deterministic identification method is used on the noisy data to obtain the final model. The method assumes the availability of a suitable computer-aided control system design package. To illustrate the method, data from a portion of a larger chemical process simulator which represents a distillation column having 2-inputs and 3-outputs is identified