Traffic Signal Time Optimization Based on Deep Q-Network

Abstract

Because cities worldwide have high population concentration, traffic congestion is a key problem that needs to be addressed. As modern technology advances, smart traffic management is able to collect data from the environment and uses a contextual signal assignment to determine the traffic flow at intersections and improve the traffic conditions. In this paper, we propose a green signal time allocation system based on a deep Q-network (DQN) that can maximize the capacity at intersections and assign the green light time according to the traffic conditions. The proposed system also aims to reduce the standard deviation of each lane at an intersection by considering the standard deviation of the waiting time. As a result, selfish green signal allocations can be reduced. Thus, the proposed system can achieve better experimental results in a dynamic environment than those of the green signal phase sequence allocation system.

Keywords

• deep Q-learning
• reinforcement learning
• traffic signal control
• capacity
• SUMO