Collaborative variable speed limit for urban expressway mainline and on-ramp in CAVs environment based on deep reinforcement learning

With the growing application of Connected and Autonomous Vehicle (CAV) technologies in active traffic management, variable speed limit (VSL) strategies are crucial for enhancing road traffic efficiency and safety. This paper addresses traffic conflicts in urban expressway merging areas, which cause reduced capacity and sudden speed changes, proposing a coordinated VSL control strategy for the mainlines and on-ramps in a Vehicle-to-Everything (V2X) environment. First, a METANET-based mainline traffic flow prediction model is used to construct a dual-objective function minimizing total travel time and distance, optimized via Model Predictive Control (MPC). Second, the VSL control problem is modeled as a Markov Decision Process (MDP) with a composite reward function based on average speed, throughput, and vehicle delay. A Deep Q-Network (DQN) algorithm is introduced to approximate the optimal on-ramp speed limits for different traffic flow states, communicated to CAVs via Infrastructure-to-Vehicle (I2V). Simulation tests on Xuzhou’s North Third Ring Expressway using the Simulation of Urban Mobility (SUMO) microscopic package were conducted to validate the strategies. Results show that the proposed method reduces total travel time by 8.51%, increases average speed by 14.49%, and lowers traffic density fluctuations by 14.81% compared to the mainline-only control. In addition, the proposed method also reduced the length of ramp queues by 37.86% and effectively reduced CO2 emissions compared to the mainline-only control. Sensitivity analysis indicates that higher CAV penetration rates significantly improve merging area efficiency and reduce speed differences between mainline and on-ramp vehicles, supporting coordinated VSL in mixed traffic on urban expressways.