Koopman operator based model predictive control of vehicle dynamics

doctoral thesis

• Title (croatian): Modelsko prediktivno upravljanje dinamikom vozila temeljeno na Koopmanovom operatoru
• Author: Marko Švec
• Mentor: Jadranko Matuško (mentor)
• Committee member: Šandor Ileš (predsjednik povjerenstva)
• Committee member: Mato Baotić (član povjerenstva)
• Committee member: Andrej Jokić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; (Department of Electric Machines, Drives and Automation); Zagreb
• Defense date and country: 2024-10-17, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering
• Universal decimal classification (UDC): 621.3 - Electrical engineering

Abstract

In recent years, the global market for electric and hybrid vehicles has grown significantly, driving technological innovation in the automotive industry. Such growth has accelerated the development of vehicle dynamics control systems, which require efficient control strategies for optimal performance. This thesis investigates the use of the Koopman operator (KO) with model predictive control (MPC) for vehicle dynamics control. The Koopman operator provides a linear representation of nonlinear systems and enables simplified control of complex dynamical systems. By integrating KO with traditional nonlinear MPC, a Koopman MPC (KMPC) framework is developed that converts nonlinear to quadratic optimization problems, improving computational efficiency with minimal loss of accuracy. Three different methods for numerical KO identification are presented: extended dynamic mode decomposition (EDMD), deep dynamic mode decomposition (Deep-DMD), and a new method called enhanced EDMD (E²DMD). A KMPC-based torque vectoring algorithm is developed and compared with existing MPC strategies in a high-fidelity simulator. Furthermore, experimental validation using a scaled vehicle model demonstrates the potential of using Koopman-based controllers in real time on a standard embedded automotive platform. The thesis also presents a new method for generating Koopman operator learning data. The results point to promising future directions in the field of vehicle dynamics control.

Abstract (croatian)

Posljednjih je godina globalno tržište električnih i hibridnih vozila značajno poraslo, potičući tako tehnološke inovacije i razvoj sustava za upravljanje dinamikom vozila, koji zahtijevaju učinkovite strategije za optimalne performanse. Ova disertacija istražuje upotrebu Koopmanovog operatora (KO) s modelskim prediktivnim upravljanjem (MPC) za upravljanje dinamikom vozila. Koopmanov operator omogućuje linearni prikaz nelinearnih sustava i pojednostavljeno upravljanje složenim dinamičkim sustavima. Integracijom s nelinearnim MPC-om razvijen je Koopman MPC (KMPC) pristup, koji pretvara nelinearne optimizacijske probleme u kvadratne i poboljšava računsku učinkovitost uz minimalni gubitak upravljačkih performansi. Predstavljene su tri metode za numeričku identifikaciju KO: proširena dinamička modalna dekompozicija (EDMD), duboka dinamička modalna dekompozicija (Deep-DMD) i nova metoda nazvana poboljšana EDMD (E²DMD). KMPC algoritam za vektoriranje zakretnog momenta uspoređen je s postojećim MPC strategijama u simulatoru visoke vjernosti, a eksperimentalna validacija na skaliranom modelu vozila pokazuje potencijal regulatora zasnovanih na Koopmanovom operatoru u stvarnom vremenu. Disertacija također predstavlja novu metodu za generiranje podataka za učenje Koopmanovog operatora. Rezultati upućuju na obećavajuće buduće smjerove u području upravljanja dinamikom vozila.

• Keywords: vehicle dynamics; model predictive control; direct yaw moment control; torque vectoring; Koopman operator; model identification; extended dynamic mode decomposition
• Keywords (croatian): dinamika vozila; modelsko prediktivno upravljanje; direktno upravljanje zakretnim momentom; vektoriranje zakretnog momenta; Koopmanov operator; identifikacija modela; proširena dinamička modalna dekompozicija
• Language: english
• URN:NBN: urn:nbn:hr:168:051099
• Promotion: 2024
• Study programme: Title: Doctoral study programme "Electrical Engineering and Computing"; Study programme type: university; Study level: postgraduate; Academic / professional title: doktor/doktorica znanosti, po-dručje tehničkih znanosti (doktor/doktorica znanosti, po-dručje tehničkih znanosti)
• Type of resource: Text
• Extent: 149 str.
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-11-28 12:54:55