Identifikacija modela vozila temeljenog na Koopmanovom operatoru korištenjem dubokih neuronskih mreža

master's thesis

• Title (english): Identification of the vehicle model based on Koopman operator using deep neural networks
• Author: Nikolina Jarak
• Mentor: Jadranko Matuško (mentor)
• Committee member: Jadranko Matuško (predsjednik povjerenstva)
• Committee member: Šandor Ileš (član povjerenstva)
• Committee member: Željko Jakopović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Model vozila predstavlja primjer dinamički nelinearmog sustava. U ovom radu opisane su duboke neuronske mreže te Koopmanov operator koji je zasnovan na algoritmu dinamičke modalne dekompozicije s ciljem identifikacije modela vozila te u
konacnosti razvijanja modela upravljanja. Prema tome, opisana je i metoda upravljanja
LQR regulatorom. Zbog složene nelinearne dinamike modela vozila, neuronska mreža
je kreirana u tri dijela: enkoder, latentni dio mreže koji sadržava Koopmanov operator te dekoder. Koopmanov operator je naucen koristeći algoritam dinamičke modalne
dekompozicije zasnovan na dubokoj neuronskoj mreži. Duboka neuronska mreža je
implementirana tako što je prvo podignuta dimenzija prostora stanja, naucen je Ko-
opmanov operator zajedno s neuronskom mrežom te su na kraju opet rekonstruirane
varijable stanja u sljedecem koraku. Podaci za učenje duboke neuronske mreže su iz-
vuceni iz CarMaker-a te je složeno sučelje unutar simulacijskog okruženja kako bi se
mogao istestirati LQR regulator. Usporedeni su dobiveni podaci iz neuronske mreže
sa stvarnim podacima te analizirani rezultati testiranja LQR regulatora.

Abstract (english)

The vehicle model is an example of a dynamically nonlinear system. In this paper deep neural networks and the Koopman operator based on the dynamical modal
decomposition algorithm with the aim of identifying vehicle model and developing
controller. Therefore, the control method with LQR regulator is also described. Due
to the comples nonlinear dynamics of the vehicle model, the neural network is created in three parts: the encoder, the latent part of the network containing the Koopman
operator and the decoder. The Koopman operator was learned using the dynamic modal decomposition algorithm based on a deep neural network. Deep neural network is
implemented by first augmenting the dimension of the state space, Koopman operator
was learned parallel with the neural network and were reconstructed again state variables in next step. Deep neural network learning data was extracted from CarMaker
and the interface was set up within the simulation environment to could test the LQR
regulator. The obtained data from the neural network were compared with actual data
and analyzed test results of LQR regulator.

• Keywords: Koopmanov operator; model vozila; duboka neuronska mreža; Deep-DMD
• Keywords (english): Koopman operator; vehicle model; deep neutral network; Deep-DMD
• Language: croatian
• URN:NBN: urn:nbn:hr:168:465583
• Study programme: Title: Electrical Engineering and Information Technology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka elektrotehnike i informacijske tehnologije (magistar/magistra inženjer/inženjerka elektrotehnike i informacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 15:08:47