Pregled bibliografske jedinice broj: 1079597
Variable Speed Limit Control Based on Deep Reinforcement Learning: A Possible Implementation
Variable Speed Limit Control Based on Deep Reinforcement Learning: A Possible Implementation // Proceedings of ELMAR-2020 / Muštra, Mario ; Vuković, Vuković ; Zovko-Cihlar, Branka (ur.).
Zagreb, 2020. str. 67-72 doi:10.1109/ELMAR49956.2020.9219031 (predavanje, međunarodna recenzija, cjeloviti rad (in extenso), znanstveni)
CROSBI ID: 1079597 Za ispravke kontaktirajte CROSBI podršku putem web obrasca
Naslov
Variable Speed Limit Control Based on Deep
Reinforcement Learning: A Possible
Implementation
Autori
Gregurić, Martin ; Kušić, Krešimir ; Vrbanić, Filip ; Ivanjko, Edouard
Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni
Izvornik
Proceedings of ELMAR-2020
/ Muštra, Mario ; Vuković, Vuković ; Zovko-Cihlar, Branka - Zagreb, 2020, 67-72
ISBN
978-1-7281-5972-0
Skup
62nd International Symposium ELMAR-2020
Mjesto i datum
Zadar, Hrvatska, 14.09.2020. - 15.09.2020
Vrsta sudjelovanja
Predavanje
Vrsta recenzije
Međunarodna recenzija
Ključne riječi
Traffic Control ; Variable Speed Limit Control ; Intelligent Transportation Systems ; Learning Systems ; Deep Q-Learning Network ; Criteria Functions ; Performance Analysis
Sažetak
Today’s urban motorways cannot fulfill their pur- pose to simultaneously serve transit and local urban trafficwith a high Level of Service. In the case of urban motorwayinfrastructure, the traditional "build only" approach is notalways possible due to the lack of space. This study is focused onthe Variable Speed Limit Control (VSLC) as one of the trafficcontrol methods applicable for any type of motorway and Q-learning as one commonly used approach for designing learningbased VSLC algorithms. The drawback of this methodology isthe representation and exploration of the large state-action spaceas it is the case in its application for VSLC. This study introducesa Deep Q-Network to approximate the Q-function and presents anovel learning approach for the VSLC application with possibilityto track vehicles on the microscopic level. The proposed rewardfunction steers the learning towards the improvement of rewardand prevention of oscillation among consecutive speed limits.
Izvorni jezik
Engleski
Znanstvena područja
Elektrotehnika, Računarstvo, Tehnologija prometa i transport
POVEZANOST RADA
Projekti:
KK.01.1.1.01.0009 - Napredne metode i tehnologije u znanosti o podatcima i kooperativnim sustavima (EK )
Ustanove:
Fakultet prometnih znanosti, Zagreb
Profili:
Martin Gregurić (autor)
Edouard Ivanjko (autor)
Filip Vrbanić (autor)
Krešimir Kušić (autor)
Citiraj ovu publikaciju:
Časopis indeksira:
- Scopus