Abstract: Selection of secondary variables is an effective way to reduce redundant information and to improve efficiency in soft sensor modeling. A novel method based on partial least-squares (PLS) regression method and false nearest neighbor (FNN) method is proposed for selecting the most suitable secondary process variables used as soft sensing inputs. In the proposed approach, the PLS regression method is employed to overcome difficulties encountered with the existing multicollinearity between the factors. In a new orthogonal space, inspired by chaos phase space FNN method, through calculation of the relativities of a certain variable in the feature subspace before and after selection, its interpretation of primary variable can be estimated, then selection of variables is carried out, and the least square method is used to obtain a soft-sensing model. This method is verified through structure test and Jolliff variable selection test, and the results demonstrate that it has good capability of secondary variable selection.

Key words: False nearest neighbor method, Feature subspace, Partial least-squares regression method, Secondary variable selection, Soft sensor modeling