Coupling response mechanism between the operational state of arterial traffic flow and carbon emissions under the goal of urban energy conservation and emission reduction

To address high emissions and low efficiency in signal control for urban arterial traffic under mixed traffic conditions, this study proposes a deep reinforcement learning signal control method oriented toward carbon emission reduction. First, a discretized state encoding matrix integrating vehicle position, speed, and acceleration is constructed to quantify the microscopic state of arterial traffic flow. Second, a multi-objective reward function that considers both fuel-vehicle carbon emissions and waiting time is designed, with dynamic weights adapted to the coupling intensity of three traffic flow states, thereby transforming the coupling mechanism into a quantifiable optimization objective. Finally, an improved deep Q-network with a dueling architecture and noisy networks is employed to optimize signal timing, smoothing traffic fluctuations and regulating the coupling relationship. Simulation results show that under free flow, stable flow, and unstable flow, average waiting times decrease to 18.3 s, 42.1 s, and 68.5 s, respectively, with a maximum reduction of 18.08%; per-vehicle CO 2 emissions drop to 142 g, 318 g, and 586 g, with a maximum reduction of 6.24%; total fuel consumption falls to 37.2 L/h, a reduction of 6.53%; and emissions of CO, HC, and NO x decline simultaneously by over 6%. Average travel time decreases to 148.9 s (an 8.26% improvement), and average travel speed increases to 35.8 km/h (a 9.15% improvement). This study effectively regulates the coupling relationship between arterial traffic flow states and carbon emissions, providing theoretical support and a technical pathway for green and low-carbon urban transportation development.