DRL-based adaptive signal control for bus priority service under connected vehicle environment

• Published in: Transportmetrica. (Abingdon, Oxfordshire, UK) Oxfordshire, UK), 2023-12, Vol.11 (1), p.1455-1477
• Main Authors: Zhang, Xinshao, He, Zhaocheng, Zhu, Yiting, You, Linlin
• Format: Article
• Language: English
• Subjects: Adaptive control; Algorithms; Buses (vehicles); deep reinforcement learning; Machine learning; Modules; Public transportation; Spatial data; Traffic intersections; traffic signal control; Traffic signals; Transit signal priority; Traveling salesman problem
• ISSN: 2168-0566; 2168-0582
• DOI: 10.1080/21680566.2023.2215955

Transit Signal Priority (TSP) strategy gives public transit vehicles privileges to pass through the intersection without stopping. Most previous studies have adopted the compulsory TSP strategy that considers to maximize the utility of public transportation, which is likely to reduce the efficiency of social vehicles. In this paper, we propose an Adaptive Transit Signal Priority (ATSP) model that considers the efficiency of both buses and social vehicles. This model has the Single Request Adaptive Transit Signal Priority (SR-ATSP) module and the Multi-Request Adaptive Transit Signal Priority (MR-ATSP) module. First, the intersection network is divided into grids based on the Discrete Traffic State Encoding (DTSE) idea to obtain the spatial information of vehicles. Then, in the SR-ATSP module, the Dueling Double Deep Q-learning Network (D3QN) algorithm is introduced to determine whether to implement the TSP strategy or not, considering the goal of minimizing the total passenger waiting time of buses and social vehicles. Based on the SR-ATSP, the MR-ATSP module introduces some rules to tackle the conflict from multiple priority requests of different buses. Simulation experiments based on an intersection in Nansha District, Guangzhou City are conducted on SUMO software. The results show that the proposed ATSP model can realize the priority treatment for of buses while reducing the waiting time of social vehicles by . It has superior performance for reducing the waiting time of buses and social vehicles than other widely-used TSP models.