Abstract
Soft sensors have been widely used for process control in industrial plants to estimate difficult-to-measure process variables online. A genetic algorithm-based process variables and dynamics selection (GAVDS) method is one method used to select important process variables and optimal time-delays of each variable simultaneously. However, the GAVDS method cannot handle a nonlinear relationship between X and an objective variable y because linear regression is used as a modeling technique. We therefore proposed a region selection method based on GAVDSand support vector regression (SVR), which is a nonlinear regression method. The proposed method is named GAVDS-SVR. We applied GAVDS-SVR to simulation data having high correlation between close pairs of X-variables and a nonlinear relationship between X and y. The GAVDS-SVR method could select regions of X-variables appropriately by considering the nonlinearity and could construct predictive models with high accuracy. Through soft-sensor analysis of industrial polymer process data, we confirmed that predictive, easy-to-interpret, and appropriate models were constructed using the proposed method.
