Locally weighted kernel partial least square model for nonlinear processes: A case study

Abstract

A soft sensor, namely locally weighted partial least squares (LW-PLS) cannot cope with the nonlinearity of process data. To address this limitation, Kernel functions are integrated into LW-PLS to form locally weighted Kernel partial least squares (LW-KPLS). In this study, the different Kernel functions including Linear Kernel, Polynomial Kernel, Exponential Kernel, Gaussian Kernel and Multiquadric Kernel were used in the LW-KPLS model. Then, the predictive performance of these Kernel functions in LW-KPLS was accessed by employing a nonlinear case study and the analysis of the obtained results was then compared. In this study, it was found that the predictive performance of using Exponential Kernel in LW-KPLS is better than other Kernel functions. The values of root-mean-square errors (RMSE) for the training and testing dataset by utilizing this Kernel function are the lowest in the case study, which is 44.54% lower RMSE values as compared to other Kernel functions.