الفهرس 
Traffic Congestion
Deep Reinforcement LearningOnly 14 pages are availabe for public view
 |  | from | 90 |  |  |
| | | from | 90 | | |
Abstract
Traffic congestion has a huge cost. Solving the traffic congestion problem has many benefits financially and environmentally. One of the methods of solving the traffic congestion problem is using a smart agent for traffic control rather than the currently deployed fixed-time traffic lights. The application of artificial intelligence to solving the traffic congestion problem has been going on for a while. However, most of the current research in this area depends on knowing lots of information about all vehicles in the network. While it produces promising results, applying these techniques in the current world is not easy.
In this research, reinforcement learning and deep reinforcement learning techniques are applied to the field of traffic control under the assumption that only minimal information is available. This approach produces results that are better than currently deployed fixed-time traffic lights without having heavy requirements. Two different models for solving the problem in a simple intersection are proposed. In the simple model test configuration, the agent’s waiting time is 82.3% of the best fixed-time traffic lights’ waiting time and the average CO2 emissions produced by the agent is 97.5% of the emissions produced by the best fixed-time traffic lights. The advanced model test configuration reaches a policy which outperforms fixed-time traffic lights, and Longest Queue First (LQF) controllers. In constant probability, the best tested configuration waiting time is 85.76% of LQF controller and 75.15% of the fixed-time controller. In random probability, the best tested configuration waiting time is 96.43% of LQF controller and 87.18% of the fixed-time controller. The advanced model is applied to a complex intersection as well and outperforms a fixed-time traffic light after a few modifications.