Variable influence on projection (VIP) for OPLS models and its applicability in multivariate time series analysis

• Abstract Recently a new parameter to infer variable importance in orthogonal projections to latent structures (OPLS) was presented. Called OPLS-VIP (variable influence on projection), this parameter is here applied in multivariate time series analysis to achieve an improved diagnosis of process dynamics. To this end, OPLS-VIP has been tested in three real-world industrial data sets; the first data set corresponds to a pulp manufacturing process using a continuous digester, the second one involves data from an industrial heater that experienced problems, and the third data set contains measures of the chemical oxygen demand into the effluent of a newsprint mill. The outcomes obtained using OPLS-VIP are benchmarked against classical PLS-VIP results. It is demonstrated how OPLS-VIP provides a better diagnosis and understanding of the time series behavior than PLS-VIP.

Ämnesord

NATURVETENSKAP  -- Kemi (hsv//swe)

NATURAL SCIENCES  -- Chemical Sciences (hsv//eng)

Nyckelord

VIP

Variable influence on projection

Multivariate time series analysis

OPLS

Variable selection

Process monitoring

Publikations- och innehållstyp

ref (ämneskategori)

art (ämneskategori)