Developing a method to provide traffic flow control to reduce fuel consumption and emissions in smart cities

Rapid urbanization, increasing population density, and the exponential growth of motor vehicle usage have significantly complicated urban transportation systems, making them less efficient and more environmentally harmful. Traffic congestion not only reduces the quality of daily urban life but also contributes to excessive fuel consumption and elevated levels of greenhouse gas emissions. Consequently, there is a growing need for sustainable, efficient, and adaptive traffic management systems. This dissertation aims to conduct a comparative evaluation of Deep Reinforcement Learning (DRL) algorithms in optimizing urban traffic flow, minimizing fuel consumption, and reducing environmental emissions. The main hypothesis of the study is that DRL algorithms with discrete action space will show higher performance and stability compared to those with continuous action space. Within the scope of this hypothesis; three scenarios with different traffic densities and signaling structures were developed, and different deep reinforcement learning algorithms were run on each scenario and their performances were compared. Performance criteria include reward points, fuel consumption and CO2 emission values; standard deviation, coefficient of variation (CV) and box plots were used in stability analyses. The findings obtained in experimental studies using DRL algorithms revealed that especially RainbowDQN and DQN algorithms are superior to other methods in terms of both performance and stability. Additionally, a conceptual comparison with the conventional Green Light Optimized Speed Advisory (GLOSA) algorithm revealed that while GLOSA performs adequately under stable conditions, it lacks adaptability in dynamic traffic environments. Overall, the study concludes that Deep Reinforcement Learning algorithms offer effective, flexible, and sustainable solutions for urban traffic control and have strong potential for integration into smart city transportation systems.