Process identification based on last principal component analysis

A simple linear identi®cation algorithm is presented in this paper. The last principal component (LPC), the eigenvector corre- sponding to the smallest eigenvalue of a non-negative symmetric matrix, contains an optimal linear relation of the column vectors of the data matrix. This traditional, well-known principal component analysis is extended to the generalized last principal compo- nent analysis (GLPC). For processes with colored measurement noise or disturbances, consistency of the GLPC estimator is achieved without involving iteration or non-linear numerical optimization. The proposed algorithm is illustrated by a simulated example and application to a pilot-scale process.