Sample selection to construct accurate locally weighted PLS models

Abstract

Recently, Just-In-Time (JIT) modeling methods, which includes locally weighted PLS, have attracted much attention because they can cope with changes in process characteristics. To construct accurate JIT models, it is very important to select an appropriate subset of samples. In past researches, samples were selected on the basis of the Euclidean distance and so on, however, the conventional methods cannot construct accurate models in some cases. In this research, a sample selection method to construct highly accurate locally weighted PLS models is developed.