A New Static Estimator Based on Self-Optimizing Theory
Abstract
This thesis compares the performance of the new static model based esti- mator proposed by Skogestad et al. (2011) with least squares (LS), principal component regression (PCR), and partial least squares (PLS) estimators on a linear, binary, and multicomponent distillation example. The performance is classified into two categories: “open-loop” performance (estimator used for monitoring) and “closed-loop” performance (estimator used for control). The new estimator is derived from a regression point of view, and it is shown that this estimator is optimal for “closed-loop” estimation. Skogestad et al. (2011) also presented a method called loss regression for applying the new estima- tor on data. This thesis shows that this estimator is sensitive to noise and collinearity, and a new improved method called the truncated "closed-loop" method (truncated CLM) is proposed. It is found that the new estimator and the truncated CLM have better “closed-loop” performance, but worse “open- loop” performance than LS, PCR and PLS.