A Study on Prevention Control of Bi-directionally Occurred Gridlock on a Single Grid Network

Abstract

Gridlock, a phenomenon in which traffic flow is severely slowed or completely stopped due to severe congestion, is anticipated to occur in urban areas after a major earthquake. In recent years, the possibility of occurring a new large-scale earthquake is believed to be relatively higher than usual in the central part of Japan. Therefore, it is essential to implement prevention control of gridlock. The author proposed the setting of cycle length and offset in order to prevent gridlock of clockwise traffic flow on a single grid network and verified its effectiveness. In this paper, we applied the result to bidirectional traffic flow, including counterclockwise traffic flow (left turns at intersections), and found its settings prevent gridlock. Furthermore, we proposed a comprehensive suppression control that combined the settings of cycle length and offset with split control based on deep reinforcement learning to maximize traffic flow, and verified its effectiveness through traffic flow simulation.