[Multi-component analysis of FTIR spectra of non-linear system using polynomial partial least squares method]

A non-linear algorithm, polynomial PLS was applied to the simultaneous analysis of OP-FTIR spectra of a five-component system whose FTIR spectra were seriously overlapped. The results were compared with the one obtained from PLS. PPLS yielded good performance, especially for the prediction of benzene and chloroform. RMSEP (root mean squared error of prediction) of benzene and chloroform in PPLS model were 0.043 and 0.087 and the corresponding values in PLS were 0.402 and 0.842, respectively. Meanwhile, variance was accounted by PPLS with fewer latent variables, which indicates the simplicity and robustness of the model. The successful application of PPLS to non-linear system was meaningful for the use of remote sensing FTIR in air monitoring.