Institutional Repository of Vilnius University
Šviesoforų valdymo optimizavimas naudojant gilųjį skatinamąjį mokymą /
Abstract
dc:descriptionThe aim of this thesis was to explore the possibilities of using deep reinforcement learning in traffic light control to achieve more efficient intersection throughput. Firstly, a literature review about reinforcement learning and artificial neural networks was made. Later on, „SUMO“ and „VISSIM“ traffic simulation softwares were compared. Finally, an experimental study was carried out to create a deep neural network for traffic light control. More precisely, an agent was developed that is able to choose a proper traffic light phase to reduce the average waiting time. The agent was trained using Q - learning algorithm. The environment in which the agent acts is a 4-way intersection created with „SUMO“. The performance of the agent was compared to the results of fixed-time traffic light control. The analysis of the simulations showed that in case of low car traffic the average waiting time can be reduced from 62 % to 71 %, in case of high car traffic - from 7 % to 34 %.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Urvikytė, Šarūnė,
- Contributors dc:contributor
-
- Valaitis, Vytautas
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD81704770&prefLang=en_US