The reinforcement learning algorithms and internet of things based control for urban traffic signal network

Adaptive traffic signal control is a challenging issue that adjusts the signal duration according to the traffic situation and manages the traffic in real-time. The performance of classical traffic signal control approaches is not sufficient to find the appropriate signal planning. In this thesis, a three-layer control method based on the Internet of Things is proposed to reduce the delay in traffic. With the proposed approach, optimization is performed at each layer. The bottom layer, the edge computing layer, provides real-time and local optimization. The middle layer, the fog computing layer, performs a real-time and global optimization process. In cloud computing, which is the top layer, the data from the lower layers are evaluated offline and parameters that will increase the performance of the lower two layers are investigated. In addition, an innovative approach is proposed to be used both in isolated intersections and in the traffic network. The methods in the literature only try to minimize the delay at intersections caused by vehicles waiting due to the red signal phase. There are three different reasons for delay at intersections: vehicles slowing down, accelerating and waiting. In this study, three different types of delay were also investigated. Proposed control uses not only the applied intersection information but also use the adjacent intersection data as an input. In this study, the SUMO traffic simulator was used for performance evaluation and method improvements. In addition, a model has been developed in which the internet of things, deep learning and reinforcement learning algorithms are used together. An architecture that can be applied both in an isolated intersection and in traffic networks formed by multiple traffic signals is designed, and it is aimed to make a great contribution to the literature and the development of intelligent traffic control system.