Use of PRESS for model structure selection

The PRESS clearly indicates that the correlations between these two inputs u2 and uz) and the process output (yi) in this particular data set are weak. However, from a physical point of view, it is expected that these two input variables, OX tower reflux flow emd temperature setpoint at the bottom of the OX tower, should have an eflfect on yi, the OX tower product purity. One possible explanation is that the settling times were overestimated for these two relationships (240 minutes each) combined with the 

During plant tests, the input variables arc sometimes adjusted by the operator in order to maintmn the product at its specification. Different corrective actions are taken in response to a particular control situation in order to avoid correlation between the independent variables. However, this type of operator intervention introduces feedback into the test data that can lead to significant bias in the estimated process models. To remove this efifect, noise 

---

USE OF PRESS FOR MODEL STRUCTURE SELECTION IN PROCESS IDENTIFICATION... 

---