Real-time prediction of product quality
or other key performance
indicators is critical to ensuring high-quality products and increasing
economic profit. In this study, a new locally weighted partial least-squares
(LW-PLS) method using optimal weighting distance-based similarity,
denoted as OLW-PLS, is proposed. OLW-PLS is a nonlinear just-in-time
modeling method, which can handle process collinearity, nonlinearity,
and time-varying characteristics. In OLW-PLS, a weighted PLS regression
model is constructed based on the optimal weighting distance-based
similarity, which considers variable interactions and nonlinear dependencies
between the input variables and the output in an optimal manner. The
feasibility and effectiveness of the proposed OLW-PLS method were
validated through its applications to a numerical example, an industrial
ethylene fractionation process, and a pharmaceutical process. The
application results have demonstrated that OLW-PLS has superior prediction
performance than the conventional PLS, LW-PLS, and CbLW-PLS.