Control of a hybrid marine propulsion plant with Reinforcement Learning

Tsiknias, Georgios; Τσικνιάς, Γεώργιος

dc.contributor.author	Tsiknias, Georgios	en
dc.contributor.author	Τσικνιάς, Γεώργιος	el
dc.date.accessioned	2021-08-31T08:04:29Z
dc.date.available	2021-08-31T08:04:29Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/53758
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.21456
dc.description	Εθνικό Μετσόβιο Πολυτεχνείο--Μεταπτυχιακή Εργασία. Διεπιστημονικό-Διατμηματικό Πρόγραμμα Μεταπτυχιακών Σπουδών (Δ.Π.Μ.Σ.) “Συστήματα Αυτοματισμού”	el
dc.rights	Default License
dc.subject	Reinforcement Learning	en
dc.subject	Control	en
dc.subject	Hybrid system	en
dc.subject	PPO	en
dc.subject	DQN	en
dc.subject	Ενισχυμένη μάθηση	el
dc.subject	'Ελεγχος	el
dc.subject	Υβριδικό σύστημα	el
dc.title	Control of a hybrid marine propulsion plant with Reinforcement Learning	en
heal.type	masterThesis
heal.classification	Έλεγχος	el
heal.classification	Controls engineering	en
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2021-06-30
heal.abstract	In this thesis, the implementation of reinforcement learning applications in a hybrid dieselelectric marine power plant is investigated. Initially, modeling procedure of the components of the power plant is been presented. For each component (engine, electric motor/ generator, battery), a models was introduced from previous studies [1]. The aim is to nd RL methods and set-ups which are accurate and computationally e cient, so that the agent would be able to solve the optimization problem in real time. Moreover, di erent environment formulations where also reviewed in order to set up a reliable simulation for the RL controller-agent. Reinforcement Learning control is a sophisticated machine learning control method which can handle nonlinear multi-variable problems with constraints by solving the optimization problem of minimizing an objective function over a nite horizon. The developed algorithms were evaluated regarding the performance with simulations in a virtual hybrid diesel-electric set-up in Julia Pluto environment and Matlab RL Designer. Finally, the performance of the developed trained agent-controllers was simulated and veri ed on working cycles with the modeled hybrid propulsion plant HIPPO-2. The simulations were conducted for various load pro les and state transitions, and the agentscontrollers were evaluated regarding the their ability to track the contextually reference and satisfy the prede ned constraints.	en
heal.advisorName	Papalambrou, George	en
heal.committeeMemberName	Papalambrou, George	en
heal.committeeMemberName	Tzafestas, Kwnstantinos	en
heal.committeeMemberName	Kyriakopoulos, Kwnstantinos	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Μηχανολόγων Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	97 σ.	el
heal.fullTextAvailability	false