Treffer: Offset-free model predictive control using Koopman-Wiener models
info:eu-repo/semantics/openAccess
© Kristian Tiiro, 2023. Except otherwise noted, the reuse of this document is authorised under a Creative Commons Attribution 4.0 International (CC-BY 4.0) licence (https://creativecommons.org/licenses/by/4.0/). This means that reuse is allowed provided appropriate credit is given and any changes are indicated. For any use or reproduction of elements that are not owned by the author(s), permission may need to be directly from the respective right holders.
English
1398428097
From OAIster®, provided by the OCLC Cooperative.
Weitere Informationen
This master’s thesis was built on the previously developed Koopman-Wiener nonlinear model predictive controller, and the goal of this thesis was to find a suitable strategy for rejecting steady-state offset, caused by plant model mismatch. This thesis also aimed to enable the controller to perform in applications where the full state is not measured and the available measurements are corrupted with noise. The work in this thesis considered multiple strategies for handling plant model mismatch, but disturbance rejection was selected as the main approach. It is proposed in this thesis that the disturbance model for disturbance rejection can be chosen by calculating empirical observability Gramian at a single initial point for every considered augmented model option and then picking the model which is interpreted as the most observable. The proposed observability analysis provides information about weak observability of the disturbance augmented model only at the single initial point. Nevertheless, it was argued in this thesis that the results can be assumed to represent the relevant operation region, and thus the method is applicable for choosing a disturbance model. As an alternative to compare against disturbance rejection, this thesis also investigated recursive least squares method that adapts the Koopman-Wiener model within the controller online. For state estimation, this thesis utilized unscented Kalman filter. This thesis demonstrated performance of the chosen methods with two nonlinear system case studies commonly studied in the literature: a simulated continuous stirred tank reactor and a simulated distillation column. This paper provides three main results. Firstly, the controller with disturbance rejection is successful in eliminating steady-state offset in a closed-loop system. Secondly, the controller is unable to reach satisfactory performance while using the recursive least squares method. Thirdly, the results from case studies support the ch