Research on Application of Deep Reinforcement Learning in Traffic Signal Control

Abstract

With the development of urbanization, traffic management has become a crucial problem. It is known that traffic signal control plays a key role in improving traffic safety and alleviating traffic pressure, which allows it to be a very popular topic in research. In recent years, deep reinforcement learning technology has been widely concerned in the field of traffic signal control since it could use deep neural network to analyze high-dimensional data and approximate the functions in reinforcement learning. This paper generally summarizes the modeling of deep reinforcement learning applying on traffic signal control. According to actual traffic conditions on the varieties of decisions that drivers make depend on run-time traffic light changes, it also provides reference for researchers in this field in the future. Besides, Action presentation using inexperience action set is put forward. Studies have found that deep reinforcement learning can well adapt to the problem of traffic signal control and comprehensively accept most of complex traffic data, but when transmitting in real word, a well-considered and practical training environment is essential.