Optimization of nonlinear multi-stage process with characteristic changes through locally-weighted partial least squares

To satisfy product quality specifications and reduce operation cost, it is crucial to optimize operating conditions of an industrial process. In this research, a new optimization method based on locally weighted partial least squares (LW-PLS) is proposed to cope with changes in process characteristics and collinearity among process variables in a nonlinear multi-stage process. To solve a nonlinear optimization problem based on justin-time modeling, self-adaptive differential evolution is adopted. The validity of the proposed method is verified through a case study, in which a manufacturing system consists of two nonlinear processes with time-varying characteristics. It is demonstrated that LW-PLS + jDE is superior to partial least squares (PLS) + sequential quadratic programmings (SQP) and kernel PLS (KPLS) + SQP.