Keywords
Deep Machine Learning, SUMO, Adaptive Traffic Light

Technologies
Python, SUMO, Q-Learning, DQN

Traffic congestion and vehicle emissions are major problems in urban traffic management. This study focuses on traffic congestion in Dublin city centre and the related carbon dioxide emissions. It examines how different traffic signal control methods affect traffic efficiency and environmental performance. Using the SUMO simulation platform, the study compares fixed-time signal control, Q-learning, and three DQN methods with different reward functions. The adaptive controller changes signal phases in real time to improve traffic flow and reduce emissions. The results show that reinforcement learning-based control performs better than fixed-time signals in several aspects. The best DQN reward model reduced carbon dioxide emissions by 25.7%, increased vehicle throughput by 5%, and reduced average waiting time by 16.5%. The comparison of different DQN reward functions also shows that reward design affects the balance between traffic efficiency and emission reduction. This study shows the potential of machine learning in intelligent traffic signal control and offers a useful reference for future low-carbon urban traffic optimisation.