Doktora Tezleri

Permanent URI for this collectionhttps://gcris.yasar.edu.tr/handle/123456789/13679

Browse

Browsing Doktora Tezleri by Author "Ağbulut, Onur"

Filter results by typing the first few letters
Now showing 1 - 1 of 1
  • Thumbnail Image
    Doctoral Thesis
    Pekiştirmeli öğrenme ile dönel kavşaklarda trafik ışıklarının optimizasyonu
    (2024) Ağbulut, Onur; Karabulut, Korhan
    Traffic congestion has been a problem as the cities and urban areas have become more crowded, which brings a rise in the number of vehicles running through roads. This increase in the number of vehicles, combined with insufficient infrastructure and the number of roads fails to cover the number of vehicles travelling every day. Some environmental effects, such as weather conditions and accidents happening on the roads also increase congestion. Traffic light controllers are the prominent elements of traffic flow in the modern world. They control the current active flow direction either by letting the vehicles pass or stop. Roundabouts are also traffic elements for regulating the traffic flow. They are an alternative traffic component to intersections that aim to organize vehicles flowing through traffic properly. In some ways, they are very similar to intersections; however, some notable differences make roundabouts distinct from intersections. Together with this, AI has made a lot of progress. Improved hardware technology made it possible to train and test AI agents much faster, thus enabling them to be applicable in a shorter time, which was very hard to achieve before. Apart from this, new solutions and algorithms have been developed that took AI and its sub-fields like ML and RL systems to become more intelligent. This also enables AI to become more engaged in the current daily life. In this thesis study, a reinforcement learning-based solution for congestion problems in traffic flow was developed to optimize roundabout traffic flow by controlling traffic lights. A solution based on the Q-Learning Algorithm is developed. Reward and state mechanisms are implemented to optimize various traffic metrics. This study aims to optimize traffic flow in roundabouts in terms of mean waiting time, the number of stopped vehicles, average speed of cars, etc., by controlling the change of traffic lights. Several experiments for different traffic flow scenarios are designed and simulated independently multiple times to obtain statistically significant results. To compare the performance of the proposed method, experiments in which traffic lights change sequentially are also designed and executed. In these experiments, outputs for different traffic metrics are calculated. Later, the results of the proposed method are compared with these baseline results. As a result, the proposed method using Q-Learning algorithm-based reinforcement learning has been observed to produce better results than baseline results in more than one metric: mean waiting time, number of stopped vehicles, average of cars, etc., especially when there is a high traffic volume.