| Popis: |
Rearranging and reparameterizing a discrete-time nonlinear model with polynomial quotient structure in input, output and parameters (xk = f(Z, p)) leads to a model linear in its (new) parameters. As a result, the parameter estimation problem becomes a so-called errors-in-variables problem for which a total least squares approach provides a natural solution. Retrieving the predictor form after estimation leads to the modified predictor: ?x = ? f(Z, ??). The objective of this paper is to evaluate the predictive quality of ?xk = ? f(Z, ??) and ?xk = f(Z, ?p) with parameters estimated using different least squares methods. The well-known Michaelis-Menten kinetics are used as an illustration with simulated (noisy) data. Finally, an example of a storage facility containing biological products is presented with real experimental data. IFAC 2006 |
