20.500.12556/RUL-83472 Spodbujevano učenje v vodenju in optimizaciji procesov Reinforcement learning in process control and optimization Podatki postajajo glavni vir 21 stoletja. Učenje in obdelava vseh teh podatkov presega sposobnosti in zmogljivosti človeka zato je uporaba strojev neizbežna. Med naborom paradigem strojnega učenja je posebej zanimivo spodbujevano učenje vendar kako se to umešča v vodenje procesov, kakšne so posebnosti, delovni okvirji, teh informacij ni na voljo. V okviru naloge smo raziskali in preučili teoretično osnovo paradigme, različne scenarije in problematike ter preizkusili in medsebojno primerjali nekatera delovna okolja. Rezultat je umestitev paradigme v področje vodenja in optimizacije ter pregled strojnega učenja na splošno. Glavni del predstavlja ključne gradnike in teoretično osnovo paradigme s pregledom glavnih algoritmov in njihovih lastnosti in tipičnih scenarijev uporabe in problematik znotraj same paradigme. Vsebinsko so predstavljane tri javno dostopne odprtokodne knjižnice in ena spletna storitev, ki kot take predstavljajo delovna in razvoja okolja. Nakazane so smernice in izhodišča za nadaljevanje študija in raziskovanja. Čeprav so algoritmi spodbujevanega učenja počasnejši v primerjavi z algoritmi v drugih paradigmah učenja, imajo širše področje uporabe in potencial za izgradnjo boljših samo učečih se strojev. Data is becoming the prime 21st century resource. Learning and processing all of this data surpasses human capability and capacity, meaning machines are unavoidable. Amongst the many machine learning paradigms, Reinforcement Learning is of especial interest; however, there is no information as to how the latter be included in process management, specifics and frameworks. Within the framework of this thesis, we researched and examined the theoretical basis for this paradigm, the various scenarios and problems, and tested and compared some of work environments, resulting in the paradigm’s inclusion in the area of processes control and optimisation, as well as providing an overview of machine learning in general. The bulk of this work presents the key building blocks and basis for the paradigm, focusing on its main algorithms and their characteristics. It also presents typical use scenarios and inherent problems within the paradigm itself. We present three public open-source libraries and one web-based service as examples of work and development environments. This thesis also presents guidelines and starting points for further study and research. Even though reinforced learning algorithms are slower when compared to other learning paradigms, they have a much wider scope of use and the potential to produce better autonomous learning machines. strojno učenje spodbujevano učenje Markovski proces odločanja funkcija vrednosti optimalna politika Machine Learning Reinforcement Learning Markov Decision Proces Value Function Optimal Policy true false false Slovenski jezik Angleški jezik Diplomsko delo 2016-06-16 08:24:07 2016-06-16 08:24:11 2022-08-01 23:22:59 0000-00-00 00:00:00 2016 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 1970-01-01 34555 Kovac_Ivan_-_Spodbujevano_ucenje_v_vodenju_in_optimizaciji_procesov.pdf Kovac_Ivan_-_Spodbujevano_ucenje_v_vodenju_in_optimizaciji_procesov.pdf 1 73AB9765DA6D05DA57090998B2990F8C 324ac301c8869a43265246b1a959f2dbfecc13f46f73cdc4abe109a8a09febe8 4847ed01-a1ae-11eb-a523-00155dcfd717 20.500.12556/rul/0af90bfd-ca48-4e09-b412-d0ce4872205c https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=86491 Fakulteta za elektrotehniko 0 0 0