Programski paket za optimalno upravljanje linearnim dinamičkim sustavima temeljem podataka

master's thesis

• Title (croatian): Software package for optimal data-based control of linear dynamical systems
• Author: Srećko Arandia-Krešić
• Mentor: Andrej Jokić (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2021-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering; Automation and Robotics

Abstract

Rad će prvotno predstaviti bihevioralni pristup teoriji upravljanja, kao polazišnu točku „data-based“ sinteze na temelju tzv. fundamentalne leme. Kao jedan od modernijih primjena bit će
naglašeno optimalno prediktivno upravljanje na temelju podataka, kao alternativa klasičnom
Modelskom prediktivnom upravljanju (eng. Model Predictive Control - MPC). Na većem broju
primjera će se pokazati sažetost identifikacije sustava na temelju fundamentalne leme, te će se
na bazi simulacija usporediti s istom temeljenom na modelu.

Sekundarno, razvit će se algoritam koji za proizvoljni diskretni linearni vremenski-invarijantni
sustav u prostoru stanja, za proizvoljnu referentnu trajektoriju, korak diskretizacije i težinske
parametre, na temelju offline podataka generira vektorski prostor svih mogućih trajektorija
sustava, te online identificira inicijalno stanja poznavanjem prethodnih izlaza, nakon kojeg
slijedi generiranje optimalne trajektorije na pomičnom horizontu - DeePC (eng. Data-enabled
predictive control) algoritam. Dodatno, u navedeni algoritam bit će integrirano i robusno
djelovanje, na temelju regularizacijskih varijabli u funkciji cilja (robustifikacija), s namjerom
da se obuhvate i nesigurnosti u identificiranom modelu, primjerice zbog utjecaja šuma. Pokazat
će se i uvjeti na garanciju performansi takvog robusnog djelovanja.

Abstract (english)

The thesis will firstly present behavioural approach to control theory, as a beginning setpoint for
data-based synthesis based on fundamental lemma. As one of the more modern approaches, the
emphasis will be on predictive optimal control based on data, as an alternative to Model
predictive control (MPC). The simplicity of system identification based on fundamental lemma
will be shown on more examples, and the comparison with model-based identification will be
made.
Secondly, the thesis shall develop an algorithm that, for a given arbitrary controllable linear
time-invariant (LTI) discrete system in state space, reference trajectory and tuning parameters,
generates the span of vector space of all possible system trajectories of a system based on offline
data, identifies an initial state based on online measurements, and produces an optimal
trajectory on finite horizon – DeePC (Data-enabled predictive control). Additionally, a robust
control based on regularization variables in cost function (robustification) will be integrated
into the algorithm, with a goal of covering the uncertainty in identified model, for example
caused by noise. The performance guarantees for such robust control will be given.

• Keywords: bihevioralni pristup; fundamentalna lema; podatci; Modelsko prediktivno upravljanje; DeePC; robustifikacija; nesigurnost
• Keywords (english): behavioural approach; fundamental lemma; data; Model predictive control; DeePC (Data-enabled predictive control); robustification; uncertainty
• Language: croatian
• URN:NBN: urn:nbn:hr:235:287502
• Study programme: Title: Mechanical Engineering; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka strojarstva (magistar/magistra inženjer/inženjerka strojarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2021-12-01 14:35:22