Performance Analysis of Deep Reinforcement Learning-based Traffic Signal Control System at Urban Bottleneck Sections

With the rapid development of transportation and urbanization, the problem of traffic congestion is becoming increasingly serious. The road congestion caused by traffic bottlenecks will spread to other areas of the road network, and alleviating traffic congestion at the bottleneck can bring improvement to the entire urban road network. By adjusting the timing of signal lights, congestion at the bottleneck can be improved to a certain extent. In response to the issue of congestion propagation caused by traffic bottlenecks, we use microscopic traffic simulation software to conduct simulation research on regional road networks with bottlenecks, and compares the performance differences between traffic signal control algorithms based on deep reinforcement learning and traditional non-learning traffic signal control algorithms. The results confirm that compared with traditional non-learning traffic signal control algorithms, traffic signal control algorithms based on deep reinforcement learning can more effectively alleviate congestion on bottleneck sections, reduce the occurrence and spread of road congestion, and have high feasibility and practicality.