Variable Speed Limit Control Based on Deep Deterministic Policy Gradient under Mixed Traffic Environment

<div>The existing variable speed limit (VSL) control strategies rely on variable message signs, leading to slow response times and sensitivity to driver compliance. These methods struggle to adapt to environments where both connected automated vehicles (CAVs) and manual vehicles coexist. This article proposes a VSL control strategy using the deep deterministic policy gradient (DDPG) algorithm to optimize travel time, reduce collision risks, and minimize energy consumption. The algorithm leverages real-time traffic data and prior speed limits to generate new control actions. A reward function is designed within a DDPG-based actor-critic framework to determine optimal speed limits. The proposed strategy was tested in two scenarios and compared against no-control, rule-based control, and DDQN-based control methods. The simulation results indicate that the proposed control strategy outperforms existing approaches in terms of improving TTS (total time spent), enhancing the throughput efficiency of the bottleneck area, and reducing the spatial and temporal extent of traffic congestion. Compared to the suboptimal DDQN-based VSL control, the proposed strategy improves TTS by 9.3% in Scenario 1 and by 11% in Scenario 2. The sensitivity analysis shows that the proposed control strategy improves performance as the penetration rate of CAVs increases. However, when the penetration rate reaches a certain threshold, the potential for further optimization becomes limited. Furthermore, higher time-to-collision (TTC) values, influenced by the reward function <i>r</i> <sub>2</sub>, enhance traffic safety.</div>