A study on the traffic signal control using the extended deep Q network

Abstract

Recently, problems such as traffic congestion and traffic accident rate are socially serious. In order to alleviate these problems, many researches on intelligent transportation systems are actively being carried out. Especially, traffic signal control method can mitigate serious traffic congestion problem through reinforcement learning model by using data such as traffic flow. Reinforcement learning algorithms evolved from deep reinforcement learning to various extensions and showed improved performance. In this paper, we applied an algorithm combining various extensions of deep reinforcement learning to traffic signal control research area. Our model learns to derive the optimal traffic signal system having the minimum traffic flow while adjusting the traffic signal length. We demonstrated that our algorithm has higher performance and faster learning speed than existing algorithms. Also, compared with the current traffic signal system, our algorithm shows the possibility of contributing to traffic alleviation when applied to real environment. © 2020 ICIC International. All rights reserved.