DDQN integrating SENet for VSL control of urban expressway mixed traffic flow

As autonomous driving technology develops, mixed traffic flow of CAVs and HVs will exist for a long time. Abundant computing resources and highly compliant CAVs in the intelligent connected environment support highway variable speed limit control, helping balance traffic flow and improve efficiency and safety. This paper proposes a deep reinforcement learning-based method: first, abstract the problem into a Markov Decision Process (MDP), integrating real-time CAV penetration and SENet-based extended merging bottleneck grid parameters into the state; use CAVs as the controller's execution carrier to indirectly control speed via interaction with HVs, then introduce DDQN to find the optimal speed limit; finally, verify the model's effectiveness through an empirical study on the merging section of Xuzhou North Third Ring Zhongshan North Road on-ramp. Results show: the DDQN model improves mixed traffic flow efficiency and environmental benefits, with multi-objective reward function outperforming single-objective; the model outperforms other comparison models, predictive factors enhance proactivity, and higher CAV penetration optimizes control effect.