Reinforcement Learning-Based Intelligent Traffic Signal Control Considering Sensing Information of Railway

Traffic signal control plays a crucial role in ensuring efficient transportation in urban road networks, and recent advancements in deep reinforcement learning (RL) have shown promising potential in this domain. However, previous studies have given limited attention to the design of the reward function while also neglecting the impact of the railway information during urban transportation. To fulfill these research gaps, this study focuses on investigating RL-based traffic signal control considering the sensing information of railway. In this study, we employ the Simulation of Urban Mobility (SUMO) software, enabling the establishment of an intelligent transportation system (ITS) field consisting of two signalized intersections and two railroad crossings. Within this simulation environment, we utilize two RL algorithms [proximal policy optimization (PPO) and deep Q-networks (DQNs)] to establish an RL-SUMO model for conducting comprehensive simulation experiments under predefined traffic conditions. Simulation results demonstrate that the proposed RL-based traffic signal control method performs significantly better compared to the fixed control. In addition, the train flow has great effects on the whole traffic flow efficiency due to the opening/closing of the railroad crossing. These findings indicate the potential of RL method in advancing traffic control strategies, paving the way for more efficient and intelligent traffic management systems.