Joint optimization of variable speed limits and locations for highway on-ramp bottlenecks

Highway on-ramps are recurrent bottlenecks to cause serious traffic delay. This study explores the joint optimization of variable speed limits (VSL) and their locations on the highway with on-ramp bottlenecks to minimize vehicle delay. To solve the proposed model, a deep reinforcement learning method based on the Deep Q-Network (DQN) based OptionCritic framework is employed to determine real-time VSL values and locations upstream and around the on-ramp bottleneck. The model is trained in an environment with random traffic volumes of 4000-6000 veh/h, enabling it to effectively handle the uncertainties of traffic flow in real-world scenarios. A case study follows to validate the effectiveness of the proposed method. The results show that the joint optimization of VSL and their locations can significantly enhance the performance of highway on-ramp bottlenecks to reduce vehicle delay, stop time, and CO₂ emissions by 24%-60%, 30%-63%, and 4%-35%, respectively. Specifically, the joint optimization of VSL and locations performs best when the traffic flow is between 5000-5500 veh/h. This research provides insights into highway on-ramp management and helps improve traffic efficiency.