Experiment design for nonlinear system identification

Abstract

Optimal experiment design has been considered as an effective tool to improve model reliability and accuracy in nonlinear system identification in the past few decades. This thesis is concerned with the following challenges which have not been previously addressed: poor initial guess problem of the nominal model in nonlinear system identification; operating points selection to improve LPV model identification accuracy; joint experimental design concerning optimal operating points and input perturbation design simultaneously. To reduce the influence of poor initial guess of a model, the proposed constrained receding-horizon design (CRHD) incorporates steady-state constraints into the design framework. The other aspect addressed is experiment design for LPV model identification. An adaptive optimal operating point design approach is developed requiring no a-priori knowledge about the true nonlinear system. Joint experiment design involving more than one experiment design factor is also considered. This problem is solved by designing the operating points and input perturbation simultaneously.