Integrating Deep Q-Networks with Rail Transit Systems for Smarter Urban Mobility

With the increasing complexity of rail transit systems, traditional signal priority strategies, which rely on fixed rules, often fail to adapt to dynamic traffic conditions and lack flexibility and scalability. This study introduces a knowledge-driven approach leveraging a deep Q-network (DQN) to optimize signal priority strategies in rail transit systems. By constructing a traffic state representation model, dynamic features are captured and input into a neural network (NN), creating a high-dimensional state space for decision-making. The DQN framework integrates an experience replay mechanism and target network updates to enhance learning stability, ensuring real-time adaptation and continuous optimization. A reward function is defined to balance train delay minimization and road-traffic coordination, achieving system-wide performance improvement. Experimental results demonstrate that, under traffic flow conditions of 300 vehicles/hour, the proposed DQN strategy reduces train delay by 50% and increases road traffic flow by 16.67%.