Identification of chemical processes using canonical variate analysis

A new identification strategy is used to estimate process models of three chemical processes. The approach is based on canonical variate analysis to select a state coordinate system that relates inputs to future outputs. Regression techniques are then used to determine a multi-input multi-output process model. The Akaike information criterion is used to determine an appropriate model order. The CVA identification methodology showed satisfactory results for the estimation of process models for three case studies-a simulated continuous stirred-tank reactor, a simulated autothermal counterflow reactor, and an experimental distillation column. The identified models were evaluated under a variety of conditions that emphasized the presence of process nonlinearities and interactions, small sample sizes relative to the process settling time, stiff dynamics, and evaluation at operating conditions different from those used for identification. The CVA process identification method is described