Estimation of influential parameters in a steady-state evaporator model: The principal component approach

Abstract

The problem solved in this paper is the identification of a steady-state computer model of a multistage countercurrent evaporator and connected cooling-tower system. Fitting the model to sets of observed input and selected output variables by the least squares principle results in a nearly-singular estimation problem. To avoid this difficulty, the principal component analysis has been used for ranking the parameters according to their influence on input—output relations. The method involves eigenvalues and eigenvectors of the approximate Fischer's information matrix computed from normalized parameter sensitivities at initial estimates of the parameters. Restricting consideration to the most influential 5 parameters out of the 10 ones originally taken into consideration, the resulting well-conditioned estimation problem has been solved successfully by a Gauss—Newton—Marquardt algorithm.