Using steady-state prior knowledge to constrain parameter estimates in nonlinear system identification

Abstract

This work investigates the use of prior knowledge in the parameter estimation of NARMAX polynomial models. The problem of parameter estimation is then formulated in such a way that the estimated models have specified features. This formulation results in a constrained optimization problem, which is solved using the ellipsoid algorithm. This technique is applied to a real DC-DC buck converter. In this system, the static relation is known from the theory but identification data are located over a rather narrow range around an operating point. Although obtained from dynamical data, the models provide good approximation to the nonlinear static function.